diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/0467b4f5-90b0-40f9-9119-6dfd78865ad4.json b/data/frontiermath/anthropic/claude-opus-4-20250514/0467b4f5-90b0-40f9-9119-6dfd78865ad4.json new file mode 100644 index 0000000000000000000000000000000000000000..9ba57b0fa9905c6a5a8cf113e645dab9d01cb5da --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/0467b4f5-90b0-40f9-9119-6dfd78865ad4.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786111930.465773", + "evaluation_timestamp": "1775558527.0", + "retrieved_timestamp": "1786111930.465773", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+8ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "OVE2", + "PLD1", + "PTA1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1875, + "uncertainty": { + "standard_error": { + "value": 0.11967838846954228 + }, + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "18", + "stopping_reason_count_repetition_guard": "17", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "500251.61", + "total_total_tokens_target_model": "9004529.0", + "avg_total_tokens_other_models": "110.50", + "total_total_tokens_other_models": "1989.0", + "avg_total_tokens_all_models": "500362.11", + "total_total_tokens_all_models": "9006518.0", + "avg_turn_count": "23.83", + "total_turn_count": "429.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/0467b4f5-90b0-40f9-9119-6dfd78865ad4_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "95fc5fd9c0a116e8d44e8bb7ec6ee2531974973ff72a96f2155c34ac2f24adbf", + "total_rows": 48 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/0467b4f5-90b0-40f9-9119-6dfd78865ad4_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/0467b4f5-90b0-40f9-9119-6dfd78865ad4_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..46111aa5dcc884956cb9d976997c58f6759c1b6a --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/0467b4f5-90b0-40f9-9119-6dfd78865ad4_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95fc5fd9c0a116e8d44e8bb7ec6ee2531974973ff72a96f2155c34ac2f24adbf +size 8134100 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/3433946a-b956-48a1-bb09-14a8f82ec93f.json b/data/frontiermath/anthropic/claude-opus-4-20250514/3433946a-b956-48a1-bb09-14a8f82ec93f.json new file mode 100644 index 0000000000000000000000000000000000000000..4cbb272cbeac1f7567f783078956967ec05ac9ea --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/3433946a-b956-48a1-bb09-14a8f82ec93f.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786113485.910628", + "evaluation_timestamp": "1775558527.0", + "retrieved_timestamp": "1786113485.910628", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+7ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.14285714285714282, + "uncertainty": { + "standard_error": { + "value": 0.06388765649999399 + }, + "num_samples": 42 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "10", + "stopping_reason_count_completed_on_successful_submit": "2", + "avg_total_tokens_target_model": "377446.83", + "total_total_tokens_target_model": "4529362.0", + "avg_total_tokens_other_models": "97.50", + "total_total_tokens_other_models": "1170.0", + "avg_total_tokens_all_models": "377544.33", + "total_total_tokens_all_models": "4530532.0", + "avg_turn_count": "22.58", + "total_turn_count": "271.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/3433946a-b956-48a1-bb09-14a8f82ec93f_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "7cb7d1d6e097fbf7fd7fbc7d9859f8913d2f65fe748c4f82fa6ec56254bc6adb", + "total_rows": 42 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/3433946a-b956-48a1-bb09-14a8f82ec93f_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/3433946a-b956-48a1-bb09-14a8f82ec93f_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d0ef1e56e70e4ba85d5f30f5eed0bf0421c02b88 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/3433946a-b956-48a1-bb09-14a8f82ec93f_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cb7d1d6e097fbf7fd7fbc7d9859f8913d2f65fe748c4f82fa6ec56254bc6adb +size 6162708 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/3491820d-0d2f-4811-b970-f186b857acf5.json b/data/frontiermath/anthropic/claude-opus-4-20250514/3491820d-0d2f-4811-b970-f186b857acf5.json new file mode 100644 index 0000000000000000000000000000000000000000..810b5c46914f6b160f3091fbcfee140e4fb48391 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/3491820d-0d2f-4811-b970-f186b857acf5.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786111879.107692", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786111879.107692", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+3ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.05555555555555555, + "uncertainty": { + "standard_error": { + "value": 0.05555555555555556 + }, + "num_samples": 18 + }, + "details": { + "total_matched_trajectories": "18", + "stopping_reason_count_repetition_guard": "17", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "453550.72", + "total_total_tokens_target_model": "8163913.0", + "avg_total_tokens_other_models": "110.50", + "total_total_tokens_other_models": "1989.0", + "avg_total_tokens_all_models": "453661.22", + "total_total_tokens_all_models": "8165902.0", + "avg_turn_count": "24.00", + "total_turn_count": "432.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/3491820d-0d2f-4811-b970-f186b857acf5_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e264a14d305909abcc58391d372814256aa45512b23cb7038a6ee46581b8c95f", + "total_rows": 18 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/3491820d-0d2f-4811-b970-f186b857acf5_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/3491820d-0d2f-4811-b970-f186b857acf5_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..7995f4fe007208437f3f956d4045572fe5970dc4 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/3491820d-0d2f-4811-b970-f186b857acf5_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e264a14d305909abcc58391d372814256aa45512b23cb7038a6ee46581b8c95f +size 2812526 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/7352a6c7-6dbe-480d-ae43-9e12bfb99ccd.json b/data/frontiermath/anthropic/claude-opus-4-20250514/7352a6c7-6dbe-480d-ae43-9e12bfb99ccd.json new file mode 100644 index 0000000000000000000000000000000000000000..125bbdcb8403c72913f9ebaf998b19af463d78d9 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/7352a6c7-6dbe-480d-ae43-9e12bfb99ccd.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786117806.998392", + "evaluation_timestamp": "1775558527.0", + "retrieved_timestamp": "1786117806.998392", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+8ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "OVE2", + "PLD1", + "PTA1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1875, + "uncertainty": { + "standard_error": { + "value": 0.11967838846954228 + }, + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "18", + "stopping_reason_count_repetition_guard": "17", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "500251.61", + "total_total_tokens_target_model": "9004529.0", + "avg_total_tokens_other_models": "110.50", + "total_total_tokens_other_models": "1989.0", + "avg_total_tokens_all_models": "500362.11", + "total_total_tokens_all_models": "9006518.0", + "avg_turn_count": "23.83", + "total_turn_count": "429.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/7352a6c7-6dbe-480d-ae43-9e12bfb99ccd_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "95fc5fd9c0a116e8d44e8bb7ec6ee2531974973ff72a96f2155c34ac2f24adbf", + "total_rows": 48 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/7352a6c7-6dbe-480d-ae43-9e12bfb99ccd_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/7352a6c7-6dbe-480d-ae43-9e12bfb99ccd_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..46111aa5dcc884956cb9d976997c58f6759c1b6a --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/7352a6c7-6dbe-480d-ae43-9e12bfb99ccd_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95fc5fd9c0a116e8d44e8bb7ec6ee2531974973ff72a96f2155c34ac2f24adbf +size 8134100 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/7a63b33b-f568-4a1f-b5e1-2a8a00541d17.json b/data/frontiermath/anthropic/claude-opus-4-20250514/7a63b33b-f568-4a1f-b5e1-2a8a00541d17.json new file mode 100644 index 0000000000000000000000000000000000000000..b9a974205113ca1aa35a5a8f90cf40976d7b4f7e --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/7a63b33b-f568-4a1f-b5e1-2a8a00541d17.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786113468.298958", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786113468.298958", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+3ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.05555555555555555, + "uncertainty": { + "standard_error": { + "value": 0.05555555555555556 + }, + "num_samples": 18 + }, + "details": { + "total_matched_trajectories": "18", + "stopping_reason_count_repetition_guard": "17", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "453550.72", + "total_total_tokens_target_model": "8163913.0", + "avg_total_tokens_other_models": "110.50", + "total_total_tokens_other_models": "1989.0", + "avg_total_tokens_all_models": "453661.22", + "total_total_tokens_all_models": "8165902.0", + "avg_turn_count": "24.00", + "total_turn_count": "432.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/7a63b33b-f568-4a1f-b5e1-2a8a00541d17_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e264a14d305909abcc58391d372814256aa45512b23cb7038a6ee46581b8c95f", + "total_rows": 18 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/7a63b33b-f568-4a1f-b5e1-2a8a00541d17_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/7a63b33b-f568-4a1f-b5e1-2a8a00541d17_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..7995f4fe007208437f3f956d4045572fe5970dc4 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/7a63b33b-f568-4a1f-b5e1-2a8a00541d17_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e264a14d305909abcc58391d372814256aa45512b23cb7038a6ee46581b8c95f +size 2812526 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/a59f3717-c5cb-4f0e-88df-bf43049dcf62.json b/data/frontiermath/anthropic/claude-opus-4-20250514/a59f3717-c5cb-4f0e-88df-bf43049dcf62.json new file mode 100644 index 0000000000000000000000000000000000000000..8b23ac09a2da235dcf84819afb5f73bdc3654882 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/a59f3717-c5cb-4f0e-88df-bf43049dcf62.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786111902.18625", + "evaluation_timestamp": "1775558527.0", + "retrieved_timestamp": "1786111902.18625", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+7ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.14285714285714282, + "uncertainty": { + "standard_error": { + "value": 0.06388765649999399 + }, + "num_samples": 42 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "10", + "stopping_reason_count_completed_on_successful_submit": "2", + "avg_total_tokens_target_model": "377446.83", + "total_total_tokens_target_model": "4529362.0", + "avg_total_tokens_other_models": "97.50", + "total_total_tokens_other_models": "1170.0", + "avg_total_tokens_all_models": "377544.33", + "total_total_tokens_all_models": "4530532.0", + "avg_turn_count": "22.58", + "total_turn_count": "271.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/a59f3717-c5cb-4f0e-88df-bf43049dcf62_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "7cb7d1d6e097fbf7fd7fbc7d9859f8913d2f65fe748c4f82fa6ec56254bc6adb", + "total_rows": 42 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/a59f3717-c5cb-4f0e-88df-bf43049dcf62_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/a59f3717-c5cb-4f0e-88df-bf43049dcf62_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d0ef1e56e70e4ba85d5f30f5eed0bf0421c02b88 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/a59f3717-c5cb-4f0e-88df-bf43049dcf62_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cb7d1d6e097fbf7fd7fbc7d9859f8913d2f65fe748c4f82fa6ec56254bc6adb +size 6162708 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/c9887853-4943-4cd6-97a3-2f5820386272.json b/data/frontiermath/anthropic/claude-opus-4-20250514/c9887853-4943-4cd6-97a3-2f5820386272.json new file mode 100644 index 0000000000000000000000000000000000000000..e96ccd72b126621f8470458ebfc5e50a060fa158 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/c9887853-4943-4cd6-97a3-2f5820386272.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786117780.414865", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786117780.414865", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+3ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.05555555555555555, + "uncertainty": { + "standard_error": { + "value": 0.05555555555555556 + }, + "num_samples": 18 + }, + "details": { + "total_matched_trajectories": "18", + "stopping_reason_count_repetition_guard": "17", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "453550.72", + "total_total_tokens_target_model": "8163913.0", + "avg_total_tokens_other_models": "110.50", + "total_total_tokens_other_models": "1989.0", + "avg_total_tokens_all_models": "453661.22", + "total_total_tokens_all_models": "8165902.0", + "avg_turn_count": "24.00", + "total_turn_count": "432.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/c9887853-4943-4cd6-97a3-2f5820386272_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e264a14d305909abcc58391d372814256aa45512b23cb7038a6ee46581b8c95f", + "total_rows": 18 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/c9887853-4943-4cd6-97a3-2f5820386272_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/c9887853-4943-4cd6-97a3-2f5820386272_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..7995f4fe007208437f3f956d4045572fe5970dc4 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/c9887853-4943-4cd6-97a3-2f5820386272_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e264a14d305909abcc58391d372814256aa45512b23cb7038a6ee46581b8c95f +size 2812526 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/e5442618-79a9-40f6-83b5-95497d0bd979.json b/data/frontiermath/anthropic/claude-opus-4-20250514/e5442618-79a9-40f6-83b5-95497d0bd979.json new file mode 100644 index 0000000000000000000000000000000000000000..3c5e824a244207b0476614a8e5893a58c1491aca --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/e5442618-79a9-40f6-83b5-95497d0bd979.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786117794.498366", + "evaluation_timestamp": "1775558527.0", + "retrieved_timestamp": "1786117794.498366", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+7ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.14285714285714282, + "uncertainty": { + "standard_error": { + "value": 0.06388765649999399 + }, + "num_samples": 42 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "10", + "stopping_reason_count_completed_on_successful_submit": "2", + "avg_total_tokens_target_model": "377446.83", + "total_total_tokens_target_model": "4529362.0", + "avg_total_tokens_other_models": "97.50", + "total_total_tokens_other_models": "1170.0", + "avg_total_tokens_all_models": "377544.33", + "total_total_tokens_all_models": "4530532.0", + "avg_turn_count": "22.58", + "total_turn_count": "271.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/e5442618-79a9-40f6-83b5-95497d0bd979_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "7cb7d1d6e097fbf7fd7fbc7d9859f8913d2f65fe748c4f82fa6ec56254bc6adb", + "total_rows": 42 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/e5442618-79a9-40f6-83b5-95497d0bd979_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/e5442618-79a9-40f6-83b5-95497d0bd979_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d0ef1e56e70e4ba85d5f30f5eed0bf0421c02b88 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/e5442618-79a9-40f6-83b5-95497d0bd979_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cb7d1d6e097fbf7fd7fbc7d9859f8913d2f65fe748c4f82fa6ec56254bc6adb +size 6162708 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/e5645312-da64-4834-87e2-9dcc3759a124.json b/data/frontiermath/anthropic/claude-opus-4-20250514/e5645312-da64-4834-87e2-9dcc3759a124.json new file mode 100644 index 0000000000000000000000000000000000000000..4cbe6134bffaa276aef07945851fb2a0ba36737a --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/e5645312-da64-4834-87e2-9dcc3759a124.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786113459.500332", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786113459.500332", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+2ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "OVE2", + "PLD1", + "PTA1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.08333333333333333, + "uncertainty": { + "standard_error": { + "value": 0.08333333333333334 + }, + "num_samples": 12 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "11", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "503489.92", + "total_total_tokens_target_model": "6041879.0", + "avg_total_tokens_other_models": "107.25", + "total_total_tokens_other_models": "1287.0", + "avg_total_tokens_all_models": "503597.17", + "total_total_tokens_all_models": "6043166.0", + "avg_turn_count": "24.42", + "total_turn_count": "293.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/e5645312-da64-4834-87e2-9dcc3759a124_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "64738dd16f2588fdfe5dd63942f851c932712f6477c95a9a0b82b376fef8d34b", + "total_rows": 12 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/e5645312-da64-4834-87e2-9dcc3759a124_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/e5645312-da64-4834-87e2-9dcc3759a124_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..93d67064140285beec9993a22b30e92d32b1e108 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/e5645312-da64-4834-87e2-9dcc3759a124_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64738dd16f2588fdfe5dd63942f851c932712f6477c95a9a0b82b376fef8d34b +size 2259562 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/e58b2c19-d015-4f82-b055-dc05e92facf3.json b/data/frontiermath/anthropic/claude-opus-4-20250514/e58b2c19-d015-4f82-b055-dc05e92facf3.json new file mode 100644 index 0000000000000000000000000000000000000000..e18291e39f90ef1442c59e4ad85c8e96ed39d3ae --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/e58b2c19-d015-4f82-b055-dc05e92facf3.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786117772.886222", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786117772.886222", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+2ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "OVE2", + "PLD1", + "PTA1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.08333333333333333, + "uncertainty": { + "standard_error": { + "value": 0.08333333333333334 + }, + "num_samples": 12 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "11", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "503489.92", + "total_total_tokens_target_model": "6041879.0", + "avg_total_tokens_other_models": "107.25", + "total_total_tokens_other_models": "1287.0", + "avg_total_tokens_all_models": "503597.17", + "total_total_tokens_all_models": "6043166.0", + "avg_turn_count": "24.42", + "total_turn_count": "293.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/e58b2c19-d015-4f82-b055-dc05e92facf3_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "64738dd16f2588fdfe5dd63942f851c932712f6477c95a9a0b82b376fef8d34b", + "total_rows": 12 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/e58b2c19-d015-4f82-b055-dc05e92facf3_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/e58b2c19-d015-4f82-b055-dc05e92facf3_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..93d67064140285beec9993a22b30e92d32b1e108 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/e58b2c19-d015-4f82-b055-dc05e92facf3_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64738dd16f2588fdfe5dd63942f851c932712f6477c95a9a0b82b376fef8d34b +size 2259562 diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/e8771ad5-5dbc-4551-8eb5-df2e6d124100.json b/data/frontiermath/anthropic/claude-opus-4-20250514/e8771ad5-5dbc-4551-8eb5-df2e6d124100.json new file mode 100644 index 0000000000000000000000000000000000000000..663fa4f589c787f9ec026f518938fb555c9bafc1 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/e8771ad5-5dbc-4551-8eb5-df2e6d124100.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-20250514/1786111865.145941", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786111865.145941", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+2ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 6, + "sample_ids": [ + "OVE2", + "PLD1", + "PTA1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.08333333333333333, + "uncertainty": { + "standard_error": { + "value": 0.08333333333333334 + }, + "num_samples": 12 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "11", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "503489.92", + "total_total_tokens_target_model": "6041879.0", + "avg_total_tokens_other_models": "107.25", + "total_total_tokens_other_models": "1287.0", + "avg_total_tokens_all_models": "503597.17", + "total_total_tokens_all_models": "6043166.0", + "avg_turn_count": "24.42", + "total_turn_count": "293.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-20250514/e8771ad5-5dbc-4551-8eb5-df2e6d124100_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "64738dd16f2588fdfe5dd63942f851c932712f6477c95a9a0b82b376fef8d34b", + "total_rows": 12 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-20250514/e8771ad5-5dbc-4551-8eb5-df2e6d124100_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-20250514/e8771ad5-5dbc-4551-8eb5-df2e6d124100_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..93d67064140285beec9993a22b30e92d32b1e108 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-20250514/e8771ad5-5dbc-4551-8eb5-df2e6d124100_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64738dd16f2588fdfe5dd63942f851c932712f6477c95a9a0b82b376fef8d34b +size 2259562 diff --git a/data/frontiermath/anthropic/claude-opus-4-6/8e1c83ac-f39a-4e04-82b3-9bf0f94b733c.json b/data/frontiermath/anthropic/claude-opus-4-6/8e1c83ac-f39a-4e04-82b3-9bf0f94b733c.json new file mode 100644 index 0000000000000000000000000000000000000000..2cf03c215c2413b4313e0183ca9f0728802e7314 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/8e1c83ac-f39a-4e04-82b3-9bf0f94b733c.json @@ -0,0 +1,100 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-6/1786117716.442563", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786117716.442563", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+1ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 3, + "sample_ids": [ + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_completed_on_successful_submit": "3", + "avg_total_tokens_target_model": "343101.33", + "total_total_tokens_target_model": "1029304.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "343101.33", + "total_total_tokens_all_models": "1029304.0", + "avg_turn_count": "10.00", + "total_turn_count": "30.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-6/8e1c83ac-f39a-4e04-82b3-9bf0f94b733c_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "5409e6d26d5a0c46094259728c7c1538abca2152e7033cb5c87a4a109e6f3cff", + "total_rows": 3 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-6/8e1c83ac-f39a-4e04-82b3-9bf0f94b733c_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-6/8e1c83ac-f39a-4e04-82b3-9bf0f94b733c_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..e5821f13ad9de48589a5d505d16769e5037525af --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/8e1c83ac-f39a-4e04-82b3-9bf0f94b733c_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5409e6d26d5a0c46094259728c7c1538abca2152e7033cb5c87a4a109e6f3cff +size 475638 diff --git a/data/frontiermath/anthropic/claude-opus-4-6/a4afc074-486d-4b87-8bcb-e24f39a3e348.json b/data/frontiermath/anthropic/claude-opus-4-6/a4afc074-486d-4b87-8bcb-e24f39a3e348.json new file mode 100644 index 0000000000000000000000000000000000000000..139b7c72255ed808eecfcee9f5658db2887405bc --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/a4afc074-486d-4b87-8bcb-e24f39a3e348.json @@ -0,0 +1,99 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-6/1786113440.939973", + "evaluation_timestamp": "1775606918.0", + "retrieved_timestamp": "1786113440.939973", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+2ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 2, + "sample_ids": [ + "CWA2", + "CWD31" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775606918.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_on_successful_submit": "4", + "avg_total_tokens_target_model": "601429.50", + "total_total_tokens_target_model": "2405718.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "601429.50", + "total_total_tokens_all_models": "2405718.0", + "avg_turn_count": "20.75", + "total_turn_count": "83.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-6/a4afc074-486d-4b87-8bcb-e24f39a3e348_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "5741f2ddca66474718ab75bd731d52ba624929e9868874603ad98267350ab8e4", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-6/a4afc074-486d-4b87-8bcb-e24f39a3e348_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-6/a4afc074-486d-4b87-8bcb-e24f39a3e348_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d432af31ec2ca5488d9b5eb2a539e7b8c11bbe8b --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/a4afc074-486d-4b87-8bcb-e24f39a3e348_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5741f2ddca66474718ab75bd731d52ba624929e9868874603ad98267350ab8e4 +size 648499 diff --git a/data/frontiermath/anthropic/claude-opus-4-6/a9d03dcf-aae3-4844-8c45-983d2f1563e1.json b/data/frontiermath/anthropic/claude-opus-4-6/a9d03dcf-aae3-4844-8c45-983d2f1563e1.json new file mode 100644 index 0000000000000000000000000000000000000000..3b2fed9a2151da26e7fb16be15335573edd9e331 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/a9d03dcf-aae3-4844-8c45-983d2f1563e1.json @@ -0,0 +1,100 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-6/1786111834.398137", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786111834.398137", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+1ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 3, + "sample_ids": [ + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_completed_on_successful_submit": "3", + "avg_total_tokens_target_model": "343101.33", + "total_total_tokens_target_model": "1029304.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "343101.33", + "total_total_tokens_all_models": "1029304.0", + "avg_turn_count": "10.00", + "total_turn_count": "30.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-6/a9d03dcf-aae3-4844-8c45-983d2f1563e1_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "5409e6d26d5a0c46094259728c7c1538abca2152e7033cb5c87a4a109e6f3cff", + "total_rows": 3 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-6/a9d03dcf-aae3-4844-8c45-983d2f1563e1_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-6/a9d03dcf-aae3-4844-8c45-983d2f1563e1_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..e5821f13ad9de48589a5d505d16769e5037525af --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/a9d03dcf-aae3-4844-8c45-983d2f1563e1_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5409e6d26d5a0c46094259728c7c1538abca2152e7033cb5c87a4a109e6f3cff +size 475638 diff --git a/data/frontiermath/anthropic/claude-opus-4-6/acc3757d-100a-40e1-8f09-fb3fb5f6fbcd.json b/data/frontiermath/anthropic/claude-opus-4-6/acc3757d-100a-40e1-8f09-fb3fb5f6fbcd.json new file mode 100644 index 0000000000000000000000000000000000000000..34eee9badd2eb601fc7bb80fceff704fbc8c47b1 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/acc3757d-100a-40e1-8f09-fb3fb5f6fbcd.json @@ -0,0 +1,100 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-6/1786113438.136655", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786113438.136655", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+1ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 3, + "sample_ids": [ + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_completed_on_successful_submit": "3", + "avg_total_tokens_target_model": "343101.33", + "total_total_tokens_target_model": "1029304.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "343101.33", + "total_total_tokens_all_models": "1029304.0", + "avg_turn_count": "10.00", + "total_turn_count": "30.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-6/acc3757d-100a-40e1-8f09-fb3fb5f6fbcd_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "5409e6d26d5a0c46094259728c7c1538abca2152e7033cb5c87a4a109e6f3cff", + "total_rows": 3 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-6/acc3757d-100a-40e1-8f09-fb3fb5f6fbcd_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-6/acc3757d-100a-40e1-8f09-fb3fb5f6fbcd_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..e5821f13ad9de48589a5d505d16769e5037525af --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/acc3757d-100a-40e1-8f09-fb3fb5f6fbcd_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5409e6d26d5a0c46094259728c7c1538abca2152e7033cb5c87a4a109e6f3cff +size 475638 diff --git a/data/frontiermath/anthropic/claude-opus-4-6/d49efcd2-4285-4d09-add9-ee82691010cf.json b/data/frontiermath/anthropic/claude-opus-4-6/d49efcd2-4285-4d09-add9-ee82691010cf.json new file mode 100644 index 0000000000000000000000000000000000000000..41a4453147519da4ea52fbea964dcc28723d4922 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/d49efcd2-4285-4d09-add9-ee82691010cf.json @@ -0,0 +1,99 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-6/1786111838.009588", + "evaluation_timestamp": "1775606918.0", + "retrieved_timestamp": "1786111838.009588", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+2ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 2, + "sample_ids": [ + "CWA2", + "CWD31" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775606918.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_on_successful_submit": "4", + "avg_total_tokens_target_model": "601429.50", + "total_total_tokens_target_model": "2405718.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "601429.50", + "total_total_tokens_all_models": "2405718.0", + "avg_turn_count": "20.75", + "total_turn_count": "83.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-6/d49efcd2-4285-4d09-add9-ee82691010cf_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "5741f2ddca66474718ab75bd731d52ba624929e9868874603ad98267350ab8e4", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-6/d49efcd2-4285-4d09-add9-ee82691010cf_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-6/d49efcd2-4285-4d09-add9-ee82691010cf_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d432af31ec2ca5488d9b5eb2a539e7b8c11bbe8b --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/d49efcd2-4285-4d09-add9-ee82691010cf_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5741f2ddca66474718ab75bd731d52ba624929e9868874603ad98267350ab8e4 +size 648499 diff --git a/data/frontiermath/anthropic/claude-opus-4-6/de31a13d-6485-4780-bde2-d20bcbb301aa.json b/data/frontiermath/anthropic/claude-opus-4-6/de31a13d-6485-4780-bde2-d20bcbb301aa.json new file mode 100644 index 0000000000000000000000000000000000000000..18e15b5f886fb2a5a08c3590520b49e6e4eee232 --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/de31a13d-6485-4780-bde2-d20bcbb301aa.json @@ -0,0 +1,99 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/anthropic_claude-opus-4-6/1786117719.359332", + "evaluation_timestamp": "1775606918.0", + "retrieved_timestamp": "1786117719.359332", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+2ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 2, + "sample_ids": [ + "CWA2", + "CWD31" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775606918.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_on_successful_submit": "4", + "avg_total_tokens_target_model": "601429.50", + "total_total_tokens_target_model": "2405718.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "601429.50", + "total_total_tokens_all_models": "2405718.0", + "avg_turn_count": "20.75", + "total_turn_count": "83.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/anthropic/claude-opus-4-6/de31a13d-6485-4780-bde2-d20bcbb301aa_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "5741f2ddca66474718ab75bd731d52ba624929e9868874603ad98267350ab8e4", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/frontiermath/anthropic/claude-opus-4-6/de31a13d-6485-4780-bde2-d20bcbb301aa_samples.jsonl b/data/frontiermath/anthropic/claude-opus-4-6/de31a13d-6485-4780-bde2-d20bcbb301aa_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d432af31ec2ca5488d9b5eb2a539e7b8c11bbe8b --- /dev/null +++ b/data/frontiermath/anthropic/claude-opus-4-6/de31a13d-6485-4780-bde2-d20bcbb301aa_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5741f2ddca66474718ab75bd731d52ba624929e9868874603ad98267350ab8e4 +size 648499 diff --git a/data/frontiermath/openai/gpt-5-2025-08-07/25b3b8b0-e40f-42d0-895e-89aff557ca79.json b/data/frontiermath/openai/gpt-5-2025-08-07/25b3b8b0-e40f-42d0-895e-89aff557ca79.json new file mode 100644 index 0000000000000000000000000000000000000000..200c545d93aa2c61fc35b6eb0f7918f5254b204b --- /dev/null +++ b/data/frontiermath/openai/gpt-5-2025-08-07/25b3b8b0-e40f-42d0-895e-89aff557ca79.json @@ -0,0 +1,116 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5-2025-08-07/1786111804.545191", + "evaluation_timestamp": "1775078458.0", + "retrieved_timestamp": "1786111804.545191", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 12, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3", + "OVE2", + "PLD1", + "PTA1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775078458.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6666666666666666, + "uncertainty": { + "standard_error": { + "value": 0.13725299209264563 + }, + "num_samples": 120 + }, + "details": { + "total_matched_trajectories": "60", + "stopping_reason_count_repetition_guard": "20", + "stopping_reason_count_completed_on_successful_submit": "40", + "avg_total_tokens_target_model": "778666.62", + "total_total_tokens_target_model": "46719997.0", + "avg_total_tokens_other_models": "39.00", + "total_total_tokens_other_models": "2340.0", + "avg_total_tokens_all_models": "778705.62", + "total_total_tokens_all_models": "46722337.0", + "avg_turn_count": "20.47", + "total_turn_count": "1228.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5-2025-08-07/25b3b8b0-e40f-42d0-895e-89aff557ca79_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "6e377cf148228f271d458b2676c1a72f1299c2a4935d2ed863b7895c453b3dfc", + "total_rows": 120 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5-2025-08-07/25b3b8b0-e40f-42d0-895e-89aff557ca79_samples.jsonl b/data/frontiermath/openai/gpt-5-2025-08-07/25b3b8b0-e40f-42d0-895e-89aff557ca79_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c65a0f903fe4aebcf3816b62a46fac00a5ca0aab --- /dev/null +++ b/data/frontiermath/openai/gpt-5-2025-08-07/25b3b8b0-e40f-42d0-895e-89aff557ca79_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e377cf148228f271d458b2676c1a72f1299c2a4935d2ed863b7895c453b3dfc +size 34029287 diff --git a/data/frontiermath/openai/gpt-5-2025-08-07/3a600726-32cb-4b1f-8dd7-26f6c60fda24.json b/data/frontiermath/openai/gpt-5-2025-08-07/3a600726-32cb-4b1f-8dd7-26f6c60fda24.json new file mode 100644 index 0000000000000000000000000000000000000000..45bf9672debb736b75f480505683c67f3fb2017f --- /dev/null +++ b/data/frontiermath/openai/gpt-5-2025-08-07/3a600726-32cb-4b1f-8dd7-26f6c60fda24.json @@ -0,0 +1,116 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5-2025-08-07/1786113417.676557", + "evaluation_timestamp": "1775078458.0", + "retrieved_timestamp": "1786113417.676557", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 12, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3", + "OVE2", + "PLD1", + "PTA1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775078458.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6666666666666666, + "uncertainty": { + "standard_error": { + "value": 0.13725299209264563 + }, + "num_samples": 120 + }, + "details": { + "total_matched_trajectories": "60", + "stopping_reason_count_repetition_guard": "20", + "stopping_reason_count_completed_on_successful_submit": "40", + "avg_total_tokens_target_model": "778666.62", + "total_total_tokens_target_model": "46719997.0", + "avg_total_tokens_other_models": "39.00", + "total_total_tokens_other_models": "2340.0", + "avg_total_tokens_all_models": "778705.62", + "total_total_tokens_all_models": "46722337.0", + "avg_turn_count": "20.47", + "total_turn_count": "1228.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5-2025-08-07/3a600726-32cb-4b1f-8dd7-26f6c60fda24_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "6e377cf148228f271d458b2676c1a72f1299c2a4935d2ed863b7895c453b3dfc", + "total_rows": 120 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5-2025-08-07/3a600726-32cb-4b1f-8dd7-26f6c60fda24_samples.jsonl b/data/frontiermath/openai/gpt-5-2025-08-07/3a600726-32cb-4b1f-8dd7-26f6c60fda24_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c65a0f903fe4aebcf3816b62a46fac00a5ca0aab --- /dev/null +++ b/data/frontiermath/openai/gpt-5-2025-08-07/3a600726-32cb-4b1f-8dd7-26f6c60fda24_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e377cf148228f271d458b2676c1a72f1299c2a4935d2ed863b7895c453b3dfc +size 34029287 diff --git a/data/frontiermath/openai/gpt-5-2025-08-07/9386221b-7497-4dfc-ab5e-5ba7658105a6.json b/data/frontiermath/openai/gpt-5-2025-08-07/9386221b-7497-4dfc-ab5e-5ba7658105a6.json new file mode 100644 index 0000000000000000000000000000000000000000..c9c9560a3581b0930dd29325aff64774eb5b707f --- /dev/null +++ b/data/frontiermath/openai/gpt-5-2025-08-07/9386221b-7497-4dfc-ab5e-5ba7658105a6.json @@ -0,0 +1,116 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5-2025-08-07/1786117686.866183", + "evaluation_timestamp": "1775078458.0", + "retrieved_timestamp": "1786117686.866183", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 12, + "sample_ids": [ + "ALL3", + "CWA2", + "CWD31", + "FMT1", + "LTI1", + "ONO3", + "OVE2", + "PLD1", + "PTA1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1775078458.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6666666666666666, + "uncertainty": { + "standard_error": { + "value": 0.13725299209264563 + }, + "num_samples": 120 + }, + "details": { + "total_matched_trajectories": "60", + "stopping_reason_count_repetition_guard": "20", + "stopping_reason_count_completed_on_successful_submit": "40", + "avg_total_tokens_target_model": "778666.62", + "total_total_tokens_target_model": "46719997.0", + "avg_total_tokens_other_models": "39.00", + "total_total_tokens_other_models": "2340.0", + "avg_total_tokens_all_models": "778705.62", + "total_total_tokens_all_models": "46722337.0", + "avg_turn_count": "20.47", + "total_turn_count": "1228.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5-2025-08-07/9386221b-7497-4dfc-ab5e-5ba7658105a6_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "6e377cf148228f271d458b2676c1a72f1299c2a4935d2ed863b7895c453b3dfc", + "total_rows": 120 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5-2025-08-07/9386221b-7497-4dfc-ab5e-5ba7658105a6_samples.jsonl b/data/frontiermath/openai/gpt-5-2025-08-07/9386221b-7497-4dfc-ab5e-5ba7658105a6_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c65a0f903fe4aebcf3816b62a46fac00a5ca0aab --- /dev/null +++ b/data/frontiermath/openai/gpt-5-2025-08-07/9386221b-7497-4dfc-ab5e-5ba7658105a6_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e377cf148228f271d458b2676c1a72f1299c2a4935d2ed863b7895c453b3dfc +size 34029287 diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/0d120bcc-ed75-4c6f-bef6-74186cfd277a.json b/data/frontiermath/openai/gpt-5.4-2026-03-05/0d120bcc-ed75-4c6f-bef6-74186cfd277a.json new file mode 100644 index 0000000000000000000000000000000000000000..55e8640d5a7b9bf236413c381a1f5db12c66247d --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/0d120bcc-ed75-4c6f-bef6-74186cfd277a.json @@ -0,0 +1,102 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5.4-2026-03-05/1786117586.200713", + "evaluation_timestamp": "1776165072.0", + "retrieved_timestamp": "1786117586.200713", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+1ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 2, + "sample_ids": [ + "CWA2", + "LTI1" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776165072.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 2 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_completed_on_successful_submit": "2", + "avg_total_tokens_target_model": "59181.00", + "total_total_tokens_target_model": "118362.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "59181.00", + "total_total_tokens_all_models": "118362.0", + "avg_turn_count": "6.50", + "total_turn_count": "13.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5.4-2026-03-05/0d120bcc-ed75-4c6f-bef6-74186cfd277a_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "bdbdb499168956ef7f82ad3d5fe2ed50109f5c4b32cb1ce2a4277fa1f1615651", + "total_rows": 2 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/0d120bcc-ed75-4c6f-bef6-74186cfd277a_samples.jsonl b/data/frontiermath/openai/gpt-5.4-2026-03-05/0d120bcc-ed75-4c6f-bef6-74186cfd277a_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..ad7c95774af3a3fa6662e8019715ea1d06cdd821 --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/0d120bcc-ed75-4c6f-bef6-74186cfd277a_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bdbdb499168956ef7f82ad3d5fe2ed50109f5c4b32cb1ce2a4277fa1f1615651 +size 151388 diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/2d5ce0c4-d254-42ef-aa30-4cae9bc2f828.json b/data/frontiermath/openai/gpt-5.4-2026-03-05/2d5ce0c4-d254-42ef-aa30-4cae9bc2f828.json new file mode 100644 index 0000000000000000000000000000000000000000..e3cb3529480f282be76da53f2194d67248a90ead --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/2d5ce0c4-d254-42ef-aa30-4cae9bc2f828.json @@ -0,0 +1,104 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5.4-2026-03-05/1786117606.714761", + "evaluation_timestamp": "1776258608.0", + "retrieved_timestamp": "1786117606.714761", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+4ep/a7843388 for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 4, + "sample_ids": [ + "CWA2", + "CWD31", + "FMT1", + "LTI1" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776258608.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 16 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_completed_on_successful_submit": "8", + "avg_total_tokens_target_model": "45446.75", + "total_total_tokens_target_model": "363574.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "45446.75", + "total_total_tokens_all_models": "363574.0", + "avg_turn_count": "6.25", + "total_turn_count": "50.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5.4-2026-03-05/2d5ce0c4-d254-42ef-aa30-4cae9bc2f828_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "b9eb3b24467ae8d559a40ff2f1f7c2db852f78c4f6dda7dc54d9d5d816121321", + "total_rows": 16 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/2d5ce0c4-d254-42ef-aa30-4cae9bc2f828_samples.jsonl b/data/frontiermath/openai/gpt-5.4-2026-03-05/2d5ce0c4-d254-42ef-aa30-4cae9bc2f828_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..2e8e2718f76261f6d502902497c3279fe584438a --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/2d5ce0c4-d254-42ef-aa30-4cae9bc2f828_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9eb3b24467ae8d559a40ff2f1f7c2db852f78c4f6dda7dc54d9d5d816121321 +size 1524208 diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/9337ec6c-8c7c-4b77-8f89-6cc789b81dae.json b/data/frontiermath/openai/gpt-5.4-2026-03-05/9337ec6c-8c7c-4b77-8f89-6cc789b81dae.json new file mode 100644 index 0000000000000000000000000000000000000000..12c7788c6bed57e4a81a245b3052a24809fae480 --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/9337ec6c-8c7c-4b77-8f89-6cc789b81dae.json @@ -0,0 +1,102 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5.4-2026-03-05/1786117592.653393", + "evaluation_timestamp": "1776165072.0", + "retrieved_timestamp": "1786117592.653393", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+2ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 2, + "sample_ids": [ + "CWD31", + "FMT1" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776165072.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_on_successful_submit": "4", + "avg_total_tokens_target_model": "154780.25", + "total_total_tokens_target_model": "619121.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "154780.25", + "total_total_tokens_all_models": "619121.0", + "avg_turn_count": "12.50", + "total_turn_count": "50.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5.4-2026-03-05/9337ec6c-8c7c-4b77-8f89-6cc789b81dae_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "ae6ff23b61fc286f033378d8de23607c291d12d3760e99b6ce59cabe814895e0", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/9337ec6c-8c7c-4b77-8f89-6cc789b81dae_samples.jsonl b/data/frontiermath/openai/gpt-5.4-2026-03-05/9337ec6c-8c7c-4b77-8f89-6cc789b81dae_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..473eb8483650b954150ed67220dbd2196abbdd9f --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/9337ec6c-8c7c-4b77-8f89-6cc789b81dae_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae6ff23b61fc286f033378d8de23607c291d12d3760e99b6ce59cabe814895e0 +size 498651 diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/b284273f-18c2-46ec-93eb-b14a942981a4.json b/data/frontiermath/openai/gpt-5.4-2026-03-05/b284273f-18c2-46ec-93eb-b14a942981a4.json new file mode 100644 index 0000000000000000000000000000000000000000..95e5f8d4352fec0930c053515a014cf1ef62930e --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/b284273f-18c2-46ec-93eb-b14a942981a4.json @@ -0,0 +1,90 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5.4-2026-03-05/1786117609.961544", + "evaluation_timestamp": "1776256372.0", + "retrieved_timestamp": "1786117609.961544", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+5ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 2, + "sample_ids": [ + "CWA2", + "LTI1" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776256372.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 10 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5.4-2026-03-05/b284273f-18c2-46ec-93eb-b14a942981a4_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "52a4c11caf6d4b354be572c60360b353dc63a3550200fad66d0f7341f577532e", + "total_rows": 10 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/b284273f-18c2-46ec-93eb-b14a942981a4_samples.jsonl b/data/frontiermath/openai/gpt-5.4-2026-03-05/b284273f-18c2-46ec-93eb-b14a942981a4_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..0330c105252854d31b82185a9159b24325d8ecea --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/b284273f-18c2-46ec-93eb-b14a942981a4_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52a4c11caf6d4b354be572c60360b353dc63a3550200fad66d0f7341f577532e +size 726925 diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/c57a5f4e-9b90-40eb-9c08-bf9057db3a2e.json b/data/frontiermath/openai/gpt-5.4-2026-03-05/c57a5f4e-9b90-40eb-9c08-bf9057db3a2e.json new file mode 100644 index 0000000000000000000000000000000000000000..a244ac279d4a55466dc8714a13b87be49eefa989 --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/c57a5f4e-9b90-40eb-9c08-bf9057db3a2e.json @@ -0,0 +1,101 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5.4-2026-03-05/1786117584.148378", + "evaluation_timestamp": "1776258608.0", + "retrieved_timestamp": "1786117584.148378", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 1, + "sample_ids": [ + "ONO3" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776258608.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_completed_on_successful_submit": "5", + "avg_total_tokens_target_model": "561340.80", + "total_total_tokens_target_model": "2806704.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "561340.80", + "total_total_tokens_all_models": "2806704.0", + "avg_turn_count": "27.20", + "total_turn_count": "136.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5.4-2026-03-05/c57a5f4e-9b90-40eb-9c08-bf9057db3a2e_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "b6bc8967af581b6970e0f9fb840c937bf16e284ef60d8e7e9d3b5b51d7789075", + "total_rows": 10 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/c57a5f4e-9b90-40eb-9c08-bf9057db3a2e_samples.jsonl b/data/frontiermath/openai/gpt-5.4-2026-03-05/c57a5f4e-9b90-40eb-9c08-bf9057db3a2e_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b0d55cc99bc821e364bde4b8e7d2c09f3ba60446 --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/c57a5f4e-9b90-40eb-9c08-bf9057db3a2e_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6bc8967af581b6970e0f9fb840c937bf16e284ef60d8e7e9d3b5b51d7789075 +size 2072553 diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/dcfd218e-42ce-4f4f-ab19-8d70559370b6.json b/data/frontiermath/openai/gpt-5.4-2026-03-05/dcfd218e-42ce-4f4f-ab19-8d70559370b6.json new file mode 100644 index 0000000000000000000000000000000000000000..5d84fcecd91eb4cba7f622225872c2fbbe278c3e --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/dcfd218e-42ce-4f4f-ab19-8d70559370b6.json @@ -0,0 +1,109 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5.4-2026-03-05/1786117614.414188", + "evaluation_timestamp": "1776165073.0", + "retrieved_timestamp": "1786117614.414188", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+7ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 5, + "sample_ids": [ + "OVE2", + "PLD1", + "RAP1", + "RSG1", + "TIK2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776165073.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.9714285714285715, + "uncertainty": { + "standard_error": { + "value": 0.028571428571428577 + }, + "num_samples": 35 + }, + "details": { + "total_matched_trajectories": "10", + "stopping_reason_count_completed_on_successful_submit": "9", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "769266.30", + "total_total_tokens_target_model": "7692663.0", + "avg_total_tokens_other_models": "11.70", + "total_total_tokens_other_models": "117.0", + "avg_total_tokens_all_models": "769278.00", + "total_total_tokens_all_models": "7692780.0", + "avg_turn_count": "13.20", + "total_turn_count": "132.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5.4-2026-03-05/dcfd218e-42ce-4f4f-ab19-8d70559370b6_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "59750a18461120a26456a565a40f6c47e38ec11ff776f9bbf498e75b7710f456", + "total_rows": 35 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/dcfd218e-42ce-4f4f-ab19-8d70559370b6_samples.jsonl b/data/frontiermath/openai/gpt-5.4-2026-03-05/dcfd218e-42ce-4f4f-ab19-8d70559370b6_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..272aa4005bfa41b20de1562b6ee0b17f676f32e5 --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/dcfd218e-42ce-4f4f-ab19-8d70559370b6_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59750a18461120a26456a565a40f6c47e38ec11ff776f9bbf498e75b7710f456 +size 5431868 diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/e722a6ff-b410-40f2-bcbe-55a2c8c7615e.json b/data/frontiermath/openai/gpt-5.4-2026-03-05/e722a6ff-b410-40f2-bcbe-55a2c8c7615e.json new file mode 100644 index 0000000000000000000000000000000000000000..e41e4f85ad81d9fda8caa95d5d9991c680f05ade --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/e722a6ff-b410-40f2-bcbe-55a2c8c7615e.json @@ -0,0 +1,102 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "frontiermath/openai_gpt-5.4-2026-03-05/1786117599.649728", + "evaluation_timestamp": "1776256372.0", + "retrieved_timestamp": "1786117599.649728", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "accuracy on frontiermath/S-adaptive+C/+4ep for scorer verification_code", + "source_data": { + "dataset_name": "frontiermath", + "source_type": "hf_dataset", + "samples_number": 2, + "sample_ids": [ + "CWD31", + "FMT1" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776256372.0", + "metric_config": { + "evaluation_description": "accuracy", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 8 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_completed_on_successful_submit": "6", + "avg_total_tokens_target_model": "111312.50", + "total_total_tokens_target_model": "667875.0", + "avg_total_tokens_other_models": "0.00", + "total_total_tokens_other_models": "0.0", + "avg_total_tokens_all_models": "111312.50", + "total_total_tokens_all_models": "667875.0", + "avg_turn_count": "11.67", + "total_turn_count": "70.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [{\"type\": \"tool\", \"name\": \"python\", \"params\": {\"timeout\": 180}}], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": [null, null, {\"type\": \"tool\", \"name\": \"inference_scaling_paper/submit_answer\", \"params\": {}}, false, \"\\n\\n\", false], \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/frontiermath/openai/gpt-5.4-2026-03-05/e722a6ff-b410-40f2-bcbe-55a2c8c7615e_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "d493f5c7487f0cca6521f673e1b7e1a6a387ef9b190e3bbd393d2ebb45e3b6df", + "total_rows": 8 + } +} \ No newline at end of file diff --git a/data/frontiermath/openai/gpt-5.4-2026-03-05/e722a6ff-b410-40f2-bcbe-55a2c8c7615e_samples.jsonl b/data/frontiermath/openai/gpt-5.4-2026-03-05/e722a6ff-b410-40f2-bcbe-55a2c8c7615e_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..fdbd83496c1aedf0a4c41dbc8b91784b2e548032 --- /dev/null +++ b/data/frontiermath/openai/gpt-5.4-2026-03-05/e722a6ff-b410-40f2-bcbe-55a2c8c7615e_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d493f5c7487f0cca6521f673e1b7e1a6a387ef9b190e3bbd393d2ebb45e3b6df +size 708948 diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/1002c3d5-267c-4e27-bba6-ea3c0f02ace3.json b/data/healthbench/anthropic/claude-opus-4-20250514/1002c3d5-267c-4e27-bba6-ea3c0f02ace3.json new file mode 100644 index 0000000000000000000000000000000000000000..f28a6d66c0ba749e609a485bc180ad6b5c7ce350 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/1002c3d5-267c-4e27-bba6-ea3c0f02ace3.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-20250514/1786118221.341918", + "evaluation_timestamp": "1776165242.0", + "retrieved_timestamp": "1786118221.341918", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+6ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "58", + "sample_ids": "c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.2824159037882528, + "uncertainty": { + "standard_deviation": 0.035423728365327256, + "num_samples": 348 + }, + "details": { + "total_matched_trajectories": "55", + "stopping_reason_count_repetition_guard": "55", + "avg_total_tokens_target_model": "12925.95", + "total_total_tokens_target_model": "710927.0", + "avg_total_tokens_other_models": "59910.58", + "total_total_tokens_other_models": "3295082.0", + "avg_total_tokens_all_models": "72836.53", + "total_total_tokens_all_models": "4006009.0", + "avg_turn_count": "31.09", + "total_turn_count": "1710.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+6ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "58", + "sample_ids": "c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.035423728365327256, + "uncertainty": { + "standard_deviation": 0.035423728365327256, + "num_samples": 348 + }, + "details": { + "total_matched_trajectories": "55", + "stopping_reason_count_repetition_guard": "55", + "avg_total_tokens_target_model": "12925.95", + "total_total_tokens_target_model": "710927.0", + "avg_total_tokens_other_models": "59910.58", + "total_total_tokens_other_models": "3295082.0", + "avg_total_tokens_all_models": "72836.53", + "total_total_tokens_all_models": "4006009.0", + "avg_turn_count": "31.09", + "total_turn_count": "1710.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+6ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "58", + "sample_ids": "c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3367198838896951, + "uncertainty": { + "standard_deviation": 0.035423728365327256, + "num_samples": 348 + }, + "details": { + "total_matched_trajectories": "55", + "stopping_reason_count_repetition_guard": "55", + "avg_total_tokens_target_model": "12925.95", + "total_total_tokens_target_model": "710927.0", + "avg_total_tokens_other_models": "59910.58", + "total_total_tokens_other_models": "3295082.0", + "avg_total_tokens_all_models": "72836.53", + "total_total_tokens_all_models": "4006009.0", + "avg_turn_count": "31.09", + "total_turn_count": "1710.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+6ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "58", + "sample_ids": "c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.2824159037882528, + "uncertainty": { + "num_samples": 348 + }, + "details": { + "total_matched_trajectories": "55", + "stopping_reason_count_repetition_guard": "55", + "avg_total_tokens_target_model": "12925.95", + "total_total_tokens_target_model": "710927.0", + "avg_total_tokens_other_models": "59910.58", + "total_total_tokens_other_models": "3295082.0", + "avg_total_tokens_all_models": "72836.53", + "total_total_tokens_all_models": "4006009.0", + "avg_turn_count": "31.09", + "total_turn_count": "1710.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-20250514/1002c3d5-267c-4e27-bba6-ea3c0f02ace3_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "cac686d9c45fec076f4516d47173f9a6195606a6d3aa441a52680af1d6db423d", + "total_rows": 348 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/1002c3d5-267c-4e27-bba6-ea3c0f02ace3_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-20250514/1002c3d5-267c-4e27-bba6-ea3c0f02ace3_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..47f097b4d274cbdcd1b7d2bcc7b8469fc535fd43 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/1002c3d5-267c-4e27-bba6-ea3c0f02ace3_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cac686d9c45fec076f4516d47173f9a6195606a6d3aa441a52680af1d6db423d +size 22808587 diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/5896dc52-aa15-4846-b85c-eaf9b1ef116b.json b/data/healthbench/anthropic/claude-opus-4-20250514/5896dc52-aa15-4846-b85c-eaf9b1ef116b.json new file mode 100644 index 0000000000000000000000000000000000000000..997a668151dbbdf897178a15823d45869c627780 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/5896dc52-aa15-4846-b85c-eaf9b1ef116b.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-20250514/1786118180.952559", + "evaluation_timestamp": "1776076028.0", + "retrieved_timestamp": "1786118180.952559", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+5ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "11", + "sample_ids": "1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed" + } + }, + "evaluation_timestamp": "1776076028.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.28708996719587554, + "uncertainty": { + "standard_deviation": 0.0804715309032432, + "num_samples": 55 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_repetition_guard": "5", + "avg_total_tokens_target_model": "10169.80", + "total_total_tokens_target_model": "50849.0", + "avg_total_tokens_other_models": "39820.80", + "total_total_tokens_other_models": "199104.0", + "avg_total_tokens_all_models": "49990.60", + "total_total_tokens_all_models": "249953.0", + "avg_turn_count": "27.00", + "total_turn_count": "135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+5ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "11", + "sample_ids": "1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed" + } + }, + "evaluation_timestamp": "1776076028.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0804715309032432, + "uncertainty": { + "standard_deviation": 0.0804715309032432, + "num_samples": 55 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_repetition_guard": "5", + "avg_total_tokens_target_model": "10169.80", + "total_total_tokens_target_model": "50849.0", + "avg_total_tokens_other_models": "39820.80", + "total_total_tokens_other_models": "199104.0", + "avg_total_tokens_all_models": "49990.60", + "total_total_tokens_all_models": "249953.0", + "avg_turn_count": "27.00", + "total_turn_count": "135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+5ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "11", + "sample_ids": "1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed" + } + }, + "evaluation_timestamp": "1776076028.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3352380952380953, + "uncertainty": { + "standard_deviation": 0.0804715309032432, + "num_samples": 55 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_repetition_guard": "5", + "avg_total_tokens_target_model": "10169.80", + "total_total_tokens_target_model": "50849.0", + "avg_total_tokens_other_models": "39820.80", + "total_total_tokens_other_models": "199104.0", + "avg_total_tokens_all_models": "49990.60", + "total_total_tokens_all_models": "249953.0", + "avg_turn_count": "27.00", + "total_turn_count": "135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+5ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "11", + "sample_ids": "1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed" + } + }, + "evaluation_timestamp": "1776076028.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.28708996719587554, + "uncertainty": { + "num_samples": 55 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_repetition_guard": "5", + "avg_total_tokens_target_model": "10169.80", + "total_total_tokens_target_model": "50849.0", + "avg_total_tokens_other_models": "39820.80", + "total_total_tokens_other_models": "199104.0", + "avg_total_tokens_all_models": "49990.60", + "total_total_tokens_all_models": "249953.0", + "avg_turn_count": "27.00", + "total_turn_count": "135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-20250514/5896dc52-aa15-4846-b85c-eaf9b1ef116b_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "fa963e9bb96215cdfead93e0d46ca3e8e01196132c98cad0f278c5e0defe0fad", + "total_rows": 55 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/5896dc52-aa15-4846-b85c-eaf9b1ef116b_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-20250514/5896dc52-aa15-4846-b85c-eaf9b1ef116b_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..3ee51d2e1e94a56450fb7488b5d5d0061e20ae71 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/5896dc52-aa15-4846-b85c-eaf9b1ef116b_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa963e9bb96215cdfead93e0d46ca3e8e01196132c98cad0f278c5e0defe0fad +size 3009355 diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/76d6571f-070a-4b24-b471-718963644fed.json b/data/healthbench/anthropic/claude-opus-4-20250514/76d6571f-070a-4b24-b471-718963644fed.json new file mode 100644 index 0000000000000000000000000000000000000000..01bde0bdd6252327a8f8bef261402b16d49f992d --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/76d6571f-070a-4b24-b471-718963644fed.json @@ -0,0 +1,322 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-20250514/1786118170.776951", + "evaluation_timestamp": "1776784035.0", + "retrieved_timestamp": "1786118170.776951", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/01742454 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "12", + "sample_ids": "99b22b39-30d5-40dc-82d9-c1cd3c793a44,3852d050-c009-4e33-b953-fd734a5b5e73,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776784035.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.09392065055623759, + "uncertainty": { + "standard_deviation": 0.06874532890279508, + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "48", + "stopping_reason_count_repetition_guard": "47", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "14290.85", + "total_total_tokens_target_model": "685961.0", + "avg_total_tokens_other_models": "67060.96", + "total_total_tokens_other_models": "3218926.0", + "avg_total_tokens_all_models": "81351.81", + "total_total_tokens_all_models": "3904887.0", + "avg_turn_count": "33.08", + "total_turn_count": "1588.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/01742454 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "12", + "sample_ids": "99b22b39-30d5-40dc-82d9-c1cd3c793a44,3852d050-c009-4e33-b953-fd734a5b5e73,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776784035.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.06874532890279508, + "uncertainty": { + "standard_deviation": 0.06874532890279508, + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "48", + "stopping_reason_count_repetition_guard": "47", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "14290.85", + "total_total_tokens_target_model": "685961.0", + "avg_total_tokens_other_models": "67060.96", + "total_total_tokens_other_models": "3218926.0", + "avg_total_tokens_all_models": "81351.81", + "total_total_tokens_all_models": "3904887.0", + "avg_turn_count": "33.08", + "total_turn_count": "1588.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/01742454 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "12", + "sample_ids": "99b22b39-30d5-40dc-82d9-c1cd3c793a44,3852d050-c009-4e33-b953-fd734a5b5e73,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776784035.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.18055555555555555, + "uncertainty": { + "standard_deviation": 0.06874532890279508, + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "48", + "stopping_reason_count_repetition_guard": "47", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "14290.85", + "total_total_tokens_target_model": "685961.0", + "avg_total_tokens_other_models": "67060.96", + "total_total_tokens_other_models": "3218926.0", + "avg_total_tokens_all_models": "81351.81", + "total_total_tokens_all_models": "3904887.0", + "avg_turn_count": "33.08", + "total_turn_count": "1588.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/01742454 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "12", + "sample_ids": "99b22b39-30d5-40dc-82d9-c1cd3c793a44,3852d050-c009-4e33-b953-fd734a5b5e73,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776784035.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.09392065055623759, + "uncertainty": { + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "48", + "stopping_reason_count_repetition_guard": "47", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "14290.85", + "total_total_tokens_target_model": "685961.0", + "avg_total_tokens_other_models": "67060.96", + "total_total_tokens_other_models": "3218926.0", + "avg_total_tokens_all_models": "81351.81", + "total_total_tokens_all_models": "3904887.0", + "avg_turn_count": "33.08", + "total_turn_count": "1588.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-20250514/76d6571f-070a-4b24-b471-718963644fed_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "9f3501a07cbdc58a5cb67ee543221f34b686035ea7ad162a86640c20707897ea", + "total_rows": 48 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/76d6571f-070a-4b24-b471-718963644fed_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-20250514/76d6571f-070a-4b24-b471-718963644fed_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..7f4025847c8f7a2e708f4f9213472ea1b8aa999e --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/76d6571f-070a-4b24-b471-718963644fed_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f3501a07cbdc58a5cb67ee543221f34b686035ea7ad162a86640c20707897ea +size 3713132 diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/a718553e-f2ca-4e88-9d8c-d9d4c452a3cd.json b/data/healthbench/anthropic/claude-opus-4-20250514/a718553e-f2ca-4e88-9d8c-d9d4c452a3cd.json new file mode 100644 index 0000000000000000000000000000000000000000..20d8cf2f78400d25d9a2d3ff5a45c50769d6063a --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/a718553e-f2ca-4e88-9d8c-d9d4c452a3cd.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-20250514/1786118164.229436", + "evaluation_timestamp": "1776779722.0", + "retrieved_timestamp": "1786118164.229436", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+3ep/dd555b6b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "54", + "sample_ids": "f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.15694530514479677, + "uncertainty": { + "standard_deviation": 0.027742803129549843, + "num_samples": 162 + }, + "details": { + "total_matched_trajectories": "162", + "stopping_reason_count_repetition_guard": "162", + "avg_total_tokens_target_model": "12218.78", + "total_total_tokens_target_model": "1979442.0", + "avg_total_tokens_other_models": "60230.96", + "total_total_tokens_other_models": "9757415.0", + "avg_total_tokens_all_models": "72449.73", + "total_total_tokens_all_models": "11736857.0", + "avg_turn_count": "30.91", + "total_turn_count": "5007.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+3ep/dd555b6b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "54", + "sample_ids": "f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.027742803129549843, + "uncertainty": { + "standard_deviation": 0.027742803129549843, + "num_samples": 162 + }, + "details": { + "total_matched_trajectories": "162", + "stopping_reason_count_repetition_guard": "162", + "avg_total_tokens_target_model": "12218.78", + "total_total_tokens_target_model": "1979442.0", + "avg_total_tokens_other_models": "60230.96", + "total_total_tokens_other_models": "9757415.0", + "avg_total_tokens_all_models": "72449.73", + "total_total_tokens_all_models": "11736857.0", + "avg_turn_count": "30.91", + "total_turn_count": "5007.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+3ep/dd555b6b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "54", + "sample_ids": "f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.2554257095158598, + "uncertainty": { + "standard_deviation": 0.027742803129549843, + "num_samples": 162 + }, + "details": { + "total_matched_trajectories": "162", + "stopping_reason_count_repetition_guard": "162", + "avg_total_tokens_target_model": "12218.78", + "total_total_tokens_target_model": "1979442.0", + "avg_total_tokens_other_models": "60230.96", + "total_total_tokens_other_models": "9757415.0", + "avg_total_tokens_all_models": "72449.73", + "total_total_tokens_all_models": "11736857.0", + "avg_turn_count": "30.91", + "total_turn_count": "5007.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+3ep/dd555b6b for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "54", + "sample_ids": "f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.15694530514479677, + "uncertainty": { + "num_samples": 162 + }, + "details": { + "total_matched_trajectories": "162", + "stopping_reason_count_repetition_guard": "162", + "avg_total_tokens_target_model": "12218.78", + "total_total_tokens_target_model": "1979442.0", + "avg_total_tokens_other_models": "60230.96", + "total_total_tokens_other_models": "9757415.0", + "avg_total_tokens_all_models": "72449.73", + "total_total_tokens_all_models": "11736857.0", + "avg_turn_count": "30.91", + "total_turn_count": "5007.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-20250514/a718553e-f2ca-4e88-9d8c-d9d4c452a3cd_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "90f66d0dec386546ba67ee9b5759c3dad7e02056dd013b86bdf392bae0a90a7d", + "total_rows": 162 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/a718553e-f2ca-4e88-9d8c-d9d4c452a3cd_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-20250514/a718553e-f2ca-4e88-9d8c-d9d4c452a3cd_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..10a71fe9f10aca3608af9d6643737cbe75a9b1fb --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/a718553e-f2ca-4e88-9d8c-d9d4c452a3cd_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90f66d0dec386546ba67ee9b5759c3dad7e02056dd013b86bdf392bae0a90a7d +size 11084326 diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/d04261fc-1264-486f-91cd-725df29c2b9f.json b/data/healthbench/anthropic/claude-opus-4-20250514/d04261fc-1264-486f-91cd-725df29c2b9f.json new file mode 100644 index 0000000000000000000000000000000000000000..fd3f3d3643aa7a6abe8d99fb7f28a223fe492a56 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/d04261fc-1264-486f-91cd-725df29c2b9f.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-20250514/1786118210.319565", + "evaluation_timestamp": "1775733996.0", + "retrieved_timestamp": "1786118210.319565", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+6ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "31", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.33437638660913, + "uncertainty": { + "standard_deviation": 0.05031422477516709, + "num_samples": 186 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "15700.55", + "total_total_tokens_target_model": "314011.0", + "avg_total_tokens_other_models": "94009.55", + "total_total_tokens_other_models": "1880191.0", + "avg_total_tokens_all_models": "109710.10", + "total_total_tokens_all_models": "2194202.0", + "avg_turn_count": "46.10", + "total_turn_count": "922.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+6ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "31", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.05031422477516709, + "uncertainty": { + "standard_deviation": 0.05031422477516709, + "num_samples": 186 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "15700.55", + "total_total_tokens_target_model": "314011.0", + "avg_total_tokens_other_models": "94009.55", + "total_total_tokens_other_models": "1880191.0", + "avg_total_tokens_all_models": "109710.10", + "total_total_tokens_all_models": "2194202.0", + "avg_turn_count": "46.10", + "total_turn_count": "922.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+6ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "31", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3388429752066116, + "uncertainty": { + "standard_deviation": 0.05031422477516709, + "num_samples": 186 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "15700.55", + "total_total_tokens_target_model": "314011.0", + "avg_total_tokens_other_models": "94009.55", + "total_total_tokens_other_models": "1880191.0", + "avg_total_tokens_all_models": "109710.10", + "total_total_tokens_all_models": "2194202.0", + "avg_turn_count": "46.10", + "total_turn_count": "922.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+6ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "31", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.33437638660913, + "uncertainty": { + "num_samples": 186 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "15700.55", + "total_total_tokens_target_model": "314011.0", + "avg_total_tokens_other_models": "94009.55", + "total_total_tokens_other_models": "1880191.0", + "avg_total_tokens_all_models": "109710.10", + "total_total_tokens_all_models": "2194202.0", + "avg_turn_count": "46.10", + "total_turn_count": "922.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-20250514/d04261fc-1264-486f-91cd-725df29c2b9f_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "b7fb1296c8ab9ac896d966de6195a0eee7d641a4f73c521ba32a379e0d787942", + "total_rows": 186 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/d04261fc-1264-486f-91cd-725df29c2b9f_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-20250514/d04261fc-1264-486f-91cd-725df29c2b9f_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..64973b6b971a29cd1831d47d75986f6635027b18 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/d04261fc-1264-486f-91cd-725df29c2b9f_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7fb1296c8ab9ac896d966de6195a0eee7d641a4f73c521ba32a379e0d787942 +size 13623847 diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/fdfd7eb6-4c0d-42eb-8ef8-bc12d7e848c8.json b/data/healthbench/anthropic/claude-opus-4-20250514/fdfd7eb6-4c0d-42eb-8ef8-bc12d7e848c8.json new file mode 100644 index 0000000000000000000000000000000000000000..66b810a4ae8a54800e518094a0662a9dcf9ff8fc --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/fdfd7eb6-4c0d-42eb-8ef8-bc12d7e848c8.json @@ -0,0 +1,314 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-20250514/1786118177.472009", + "evaluation_timestamp": "1776696116.0", + "retrieved_timestamp": "1786118177.472009", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-20250514", + "id": "anthropic/claude-opus-4-20250514", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/3523d247 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "18", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1417977508618592, + "uncertainty": { + "standard_deviation": 0.03209772787538795, + "num_samples": 72 + }, + "details": { + "total_matched_trajectories": "72", + "stopping_reason_count_repetition_guard": "72", + "avg_total_tokens_target_model": "13637.58", + "total_total_tokens_target_model": "981906.0", + "avg_total_tokens_other_models": "72809.07", + "total_total_tokens_other_models": "5242253.0", + "avg_total_tokens_all_models": "86446.65", + "total_total_tokens_all_models": "6224159.0", + "avg_turn_count": "38.81", + "total_turn_count": "2794.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/3523d247 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "18", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.03209772787538795, + "uncertainty": { + "standard_deviation": 0.03209772787538795, + "num_samples": 72 + }, + "details": { + "total_matched_trajectories": "72", + "stopping_reason_count_repetition_guard": "72", + "avg_total_tokens_target_model": "13637.58", + "total_total_tokens_target_model": "981906.0", + "avg_total_tokens_other_models": "72809.07", + "total_total_tokens_other_models": "5242253.0", + "avg_total_tokens_all_models": "86446.65", + "total_total_tokens_all_models": "6224159.0", + "avg_turn_count": "38.81", + "total_turn_count": "2794.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/3523d247 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "18", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.228125, + "uncertainty": { + "standard_deviation": 0.03209772787538795, + "num_samples": 72 + }, + "details": { + "total_matched_trajectories": "72", + "stopping_reason_count_repetition_guard": "72", + "avg_total_tokens_target_model": "13637.58", + "total_total_tokens_target_model": "981906.0", + "avg_total_tokens_other_models": "72809.07", + "total_total_tokens_other_models": "5242253.0", + "avg_total_tokens_all_models": "86446.65", + "total_total_tokens_all_models": "6224159.0", + "avg_turn_count": "38.81", + "total_turn_count": "2794.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/3523d247 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "18", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1417977508618592, + "uncertainty": { + "num_samples": 72 + }, + "details": { + "total_matched_trajectories": "72", + "stopping_reason_count_repetition_guard": "72", + "avg_total_tokens_target_model": "13637.58", + "total_total_tokens_target_model": "981906.0", + "avg_total_tokens_other_models": "72809.07", + "total_total_tokens_other_models": "5242253.0", + "avg_total_tokens_all_models": "86446.65", + "total_total_tokens_all_models": "6224159.0", + "avg_turn_count": "38.81", + "total_turn_count": "2794.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-20250514/fdfd7eb6-4c0d-42eb-8ef8-bc12d7e848c8_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "96e6bffc1f00c4a040fd77f594a6b4ef2376d1b790e294254ac874f2c61ca2ad", + "total_rows": 72 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-20250514/fdfd7eb6-4c0d-42eb-8ef8-bc12d7e848c8_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-20250514/fdfd7eb6-4c0d-42eb-8ef8-bc12d7e848c8_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..31546d81e406b826339deee81bd03ae6b505f084 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-20250514/fdfd7eb6-4c0d-42eb-8ef8-bc12d7e848c8_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96e6bffc1f00c4a040fd77f594a6b4ef2376d1b790e294254ac874f2c61ca2ad +size 5867513 diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/31f22d53-950b-4347-bd1f-f6452b8be826.json b/data/healthbench/anthropic/claude-opus-4-5-20251101/31f22d53-950b-4347-bd1f-f6452b8be826.json new file mode 100644 index 0000000000000000000000000000000000000000..f81516f40afc4bf1f0ce5b9f4bef0d37a6b72b5d --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/31f22d53-950b-4347-bd1f-f6452b8be826.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-5-20251101/1786118069.119746", + "evaluation_timestamp": "1776335705.0", + "retrieved_timestamp": "1786118069.119746", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-5-20251101", + "id": "anthropic/claude-opus-4-5-20251101", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+3ep/e4e2a7fc for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "16", + "sample_ids": "ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.32475849915782595, + "uncertainty": { + "standard_deviation": 0.07011996938064514, + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "12807.67", + "total_total_tokens_target_model": "153692.0", + "avg_total_tokens_other_models": "52618.17", + "total_total_tokens_other_models": "631418.0", + "avg_total_tokens_all_models": "65425.83", + "total_total_tokens_all_models": "785110.0", + "avg_turn_count": "34.33", + "total_turn_count": "412.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+3ep/e4e2a7fc for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "16", + "sample_ids": "ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.07011996938064514, + "uncertainty": { + "standard_deviation": 0.07011996938064514, + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "12807.67", + "total_total_tokens_target_model": "153692.0", + "avg_total_tokens_other_models": "52618.17", + "total_total_tokens_other_models": "631418.0", + "avg_total_tokens_all_models": "65425.83", + "total_total_tokens_all_models": "785110.0", + "avg_turn_count": "34.33", + "total_turn_count": "412.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+3ep/e4e2a7fc for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "16", + "sample_ids": "ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3160919540229885, + "uncertainty": { + "standard_deviation": 0.07011996938064514, + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "12807.67", + "total_total_tokens_target_model": "153692.0", + "avg_total_tokens_other_models": "52618.17", + "total_total_tokens_other_models": "631418.0", + "avg_total_tokens_all_models": "65425.83", + "total_total_tokens_all_models": "785110.0", + "avg_turn_count": "34.33", + "total_turn_count": "412.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+3ep/e4e2a7fc for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "16", + "sample_ids": "ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.32475849915782595, + "uncertainty": { + "num_samples": 48 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "12807.67", + "total_total_tokens_target_model": "153692.0", + "avg_total_tokens_other_models": "52618.17", + "total_total_tokens_other_models": "631418.0", + "avg_total_tokens_all_models": "65425.83", + "total_total_tokens_all_models": "785110.0", + "avg_turn_count": "34.33", + "total_turn_count": "412.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-5-20251101/31f22d53-950b-4347-bd1f-f6452b8be826_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "45721c6aff0208d5f145e2c013b2f972d55eb480e3d9e1c443cf7869e88ad94f", + "total_rows": 48 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/31f22d53-950b-4347-bd1f-f6452b8be826_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-5-20251101/31f22d53-950b-4347-bd1f-f6452b8be826_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d50cd7c19b0586eae4e5f19e7807dc7cf287e14f --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/31f22d53-950b-4347-bd1f-f6452b8be826_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45721c6aff0208d5f145e2c013b2f972d55eb480e3d9e1c443cf7869e88ad94f +size 2818952 diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/35cf78a8-fe7c-4b9b-9dd7-5bcc784f6dc6.json b/data/healthbench/anthropic/claude-opus-4-5-20251101/35cf78a8-fe7c-4b9b-9dd7-5bcc784f6dc6.json new file mode 100644 index 0000000000000000000000000000000000000000..4dffb90eae87b7138361494ac75077f668dd8a28 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/35cf78a8-fe7c-4b9b-9dd7-5bcc784f6dc6.json @@ -0,0 +1,314 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-5-20251101/1786118105.913496", + "evaluation_timestamp": "1776696116.0", + "retrieved_timestamp": "1786118105.913496", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-5-20251101", + "id": "anthropic/claude-opus-4-5-20251101", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/1e600f58 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,59c68d10-55c1-4629-b05a-bedde7f343b5" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1614795477802404, + "uncertainty": { + "standard_deviation": 0.03178647374076512, + "num_samples": 200 + }, + "details": { + "total_matched_trajectories": "200", + "stopping_reason_count_repetition_guard": "200", + "avg_total_tokens_target_model": "20100.06", + "total_total_tokens_target_model": "4020011.0", + "avg_total_tokens_other_models": "62567.00", + "total_total_tokens_other_models": "12513399.0", + "avg_total_tokens_all_models": "82667.05", + "total_total_tokens_all_models": "16533410.0", + "avg_turn_count": "34.94", + "total_turn_count": "6987.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/1e600f58 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,59c68d10-55c1-4629-b05a-bedde7f343b5" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.03178647374076512, + "uncertainty": { + "standard_deviation": 0.03178647374076512, + "num_samples": 200 + }, + "details": { + "total_matched_trajectories": "200", + "stopping_reason_count_repetition_guard": "200", + "avg_total_tokens_target_model": "20100.06", + "total_total_tokens_target_model": "4020011.0", + "avg_total_tokens_other_models": "62567.00", + "total_total_tokens_other_models": "12513399.0", + "avg_total_tokens_all_models": "82667.05", + "total_total_tokens_all_models": "16533410.0", + "avg_turn_count": "34.94", + "total_turn_count": "6987.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/1e600f58 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,59c68d10-55c1-4629-b05a-bedde7f343b5" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.23106060606060605, + "uncertainty": { + "standard_deviation": 0.03178647374076512, + "num_samples": 200 + }, + "details": { + "total_matched_trajectories": "200", + "stopping_reason_count_repetition_guard": "200", + "avg_total_tokens_target_model": "20100.06", + "total_total_tokens_target_model": "4020011.0", + "avg_total_tokens_other_models": "62567.00", + "total_total_tokens_other_models": "12513399.0", + "avg_total_tokens_all_models": "82667.05", + "total_total_tokens_all_models": "16533410.0", + "avg_turn_count": "34.94", + "total_turn_count": "6987.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/1e600f58 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,59c68d10-55c1-4629-b05a-bedde7f343b5" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1614795477802404, + "uncertainty": { + "num_samples": 200 + }, + "details": { + "total_matched_trajectories": "200", + "stopping_reason_count_repetition_guard": "200", + "avg_total_tokens_target_model": "20100.06", + "total_total_tokens_target_model": "4020011.0", + "avg_total_tokens_other_models": "62567.00", + "total_total_tokens_other_models": "12513399.0", + "avg_total_tokens_all_models": "82667.05", + "total_total_tokens_all_models": "16533410.0", + "avg_turn_count": "34.94", + "total_turn_count": "6987.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-5-20251101/35cf78a8-fe7c-4b9b-9dd7-5bcc784f6dc6_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "fb5edd0f7058350bb36b2f1dd162ad0c39dce41638f215c68ffbe1035793e752", + "total_rows": 200 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/35cf78a8-fe7c-4b9b-9dd7-5bcc784f6dc6_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-5-20251101/35cf78a8-fe7c-4b9b-9dd7-5bcc784f6dc6_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..dfc21a4b1acc9cbd1c895bbe4e57df879d8db742 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/35cf78a8-fe7c-4b9b-9dd7-5bcc784f6dc6_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb5edd0f7058350bb36b2f1dd162ad0c39dce41638f215c68ffbe1035793e752 +size 14776679 diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/74912f4c-ebef-4f02-8e22-7327eb0f224f.json b/data/healthbench/anthropic/claude-opus-4-5-20251101/74912f4c-ebef-4f02-8e22-7327eb0f224f.json new file mode 100644 index 0000000000000000000000000000000000000000..c4cb158484a15440fcfd04ca09059ff456a8900b --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/74912f4c-ebef-4f02-8e22-7327eb0f224f.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-5-20251101/1786118066.213725", + "evaluation_timestamp": "1776335705.0", + "retrieved_timestamp": "1786118066.213725", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-5-20251101", + "id": "anthropic/claude-opus-4-5-20251101", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep/eea1466c for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.01465201465201467, + "uncertainty": { + "standard_deviation": 0.09912448378393966, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "10409.50", + "total_total_tokens_target_model": "20819.0", + "avg_total_tokens_other_models": "91104.00", + "total_total_tokens_other_models": "182208.0", + "avg_total_tokens_all_models": "101513.50", + "total_total_tokens_all_models": "203027.0", + "avg_turn_count": "39.00", + "total_turn_count": "78.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep/eea1466c for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.09912448378393966, + "uncertainty": { + "standard_deviation": 0.09912448378393966, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "10409.50", + "total_total_tokens_target_model": "20819.0", + "avg_total_tokens_other_models": "91104.00", + "total_total_tokens_other_models": "182208.0", + "avg_total_tokens_all_models": "101513.50", + "total_total_tokens_all_models": "203027.0", + "avg_turn_count": "39.00", + "total_turn_count": "78.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep/eea1466c for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.25, + "uncertainty": { + "standard_deviation": 0.09912448378393966, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "10409.50", + "total_total_tokens_target_model": "20819.0", + "avg_total_tokens_other_models": "91104.00", + "total_total_tokens_other_models": "182208.0", + "avg_total_tokens_all_models": "101513.50", + "total_total_tokens_all_models": "203027.0", + "avg_turn_count": "39.00", + "total_turn_count": "78.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep/eea1466c for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.01465201465201467, + "uncertainty": { + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "10409.50", + "total_total_tokens_target_model": "20819.0", + "avg_total_tokens_other_models": "91104.00", + "total_total_tokens_other_models": "182208.0", + "avg_total_tokens_all_models": "101513.50", + "total_total_tokens_all_models": "203027.0", + "avg_turn_count": "39.00", + "total_turn_count": "78.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-5-20251101/74912f4c-ebef-4f02-8e22-7327eb0f224f_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "d42bc6baaf3e0bf21863096f1964a3efb165a84ce6123e04256b56d728feb2f0", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/74912f4c-ebef-4f02-8e22-7327eb0f224f_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-5-20251101/74912f4c-ebef-4f02-8e22-7327eb0f224f_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..ac10aba4bbd0f983eb5bc2deab2b96b03d35b960 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/74912f4c-ebef-4f02-8e22-7327eb0f224f_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d42bc6baaf3e0bf21863096f1964a3efb165a84ce6123e04256b56d728feb2f0 +size 362344 diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/937dc864-d7a5-4d3a-a9e6-db052a70d3f7.json b/data/healthbench/anthropic/claude-opus-4-5-20251101/937dc864-d7a5-4d3a-a9e6-db052a70d3f7.json new file mode 100644 index 0000000000000000000000000000000000000000..656ed3f302e9833882c7f1b24f9851e7ea55cccb --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/937dc864-d7a5-4d3a-a9e6-db052a70d3f7.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-5-20251101/1786118075.679734", + "evaluation_timestamp": "1776335705.0", + "retrieved_timestamp": "1786118075.679734", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-5-20251101", + "id": "anthropic/claude-opus-4-5-20251101", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/8665ccea for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "74a50705-7db2-44be-84dc-92b2d960d344,48669cdc-1a64-4c99-a6e8-9a7181341fa2" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.07099885408708939, + "uncertainty": { + "standard_deviation": 0.0550047569473184, + "num_samples": 8 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "12158.75", + "total_total_tokens_target_model": "48635.0", + "avg_total_tokens_other_models": "39846.50", + "total_total_tokens_other_models": "159386.0", + "avg_total_tokens_all_models": "52005.25", + "total_total_tokens_all_models": "208021.0", + "avg_turn_count": "28.25", + "total_turn_count": "113.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/8665ccea for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "74a50705-7db2-44be-84dc-92b2d960d344,48669cdc-1a64-4c99-a6e8-9a7181341fa2" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0550047569473184, + "uncertainty": { + "standard_deviation": 0.0550047569473184, + "num_samples": 8 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "12158.75", + "total_total_tokens_target_model": "48635.0", + "avg_total_tokens_other_models": "39846.50", + "total_total_tokens_other_models": "159386.0", + "avg_total_tokens_all_models": "52005.25", + "total_total_tokens_all_models": "208021.0", + "avg_turn_count": "28.25", + "total_turn_count": "113.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/8665ccea for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "74a50705-7db2-44be-84dc-92b2d960d344,48669cdc-1a64-4c99-a6e8-9a7181341fa2" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.2804878048780488, + "uncertainty": { + "standard_deviation": 0.0550047569473184, + "num_samples": 8 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "12158.75", + "total_total_tokens_target_model": "48635.0", + "avg_total_tokens_other_models": "39846.50", + "total_total_tokens_other_models": "159386.0", + "avg_total_tokens_all_models": "52005.25", + "total_total_tokens_all_models": "208021.0", + "avg_turn_count": "28.25", + "total_turn_count": "113.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/8665ccea for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "74a50705-7db2-44be-84dc-92b2d960d344,48669cdc-1a64-4c99-a6e8-9a7181341fa2" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.07099885408708939, + "uncertainty": { + "num_samples": 8 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "12158.75", + "total_total_tokens_target_model": "48635.0", + "avg_total_tokens_other_models": "39846.50", + "total_total_tokens_other_models": "159386.0", + "avg_total_tokens_all_models": "52005.25", + "total_total_tokens_all_models": "208021.0", + "avg_turn_count": "28.25", + "total_turn_count": "113.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-5-20251101/937dc864-d7a5-4d3a-a9e6-db052a70d3f7_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e06b93c0587fd74f56a1b04df28e40689f8db4f0a6c3a64f11f6e0d7b375f771", + "total_rows": 8 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/937dc864-d7a5-4d3a-a9e6-db052a70d3f7_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-5-20251101/937dc864-d7a5-4d3a-a9e6-db052a70d3f7_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..eedff97c62401507a308e49b191ffc7c2e017e1b --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/937dc864-d7a5-4d3a-a9e6-db052a70d3f7_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e06b93c0587fd74f56a1b04df28e40689f8db4f0a6c3a64f11f6e0d7b375f771 +size 1128443 diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/caaf59c0-f77d-4878-949c-bc57d8e49cb2.json b/data/healthbench/anthropic/claude-opus-4-5-20251101/caaf59c0-f77d-4878-949c-bc57d8e49cb2.json new file mode 100644 index 0000000000000000000000000000000000000000..c192f479dfcdee89ea516a0df79be51ca1cf6734 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/caaf59c0-f77d-4878-949c-bc57d8e49cb2.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-5-20251101/1786118064.505721", + "evaluation_timestamp": "1776779722.0", + "retrieved_timestamp": "1786118064.505721", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-5-20251101", + "id": "anthropic/claude-opus-4-5-20251101", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+1ep/4a160ca4 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "54349c88-e80a-4728-b181-305f97905d59,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0, + "uncertainty": { + "standard_deviation": 0.09240718108780466, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "10", + "stopping_reason_count_repetition_guard": "10", + "avg_total_tokens_target_model": "18758.10", + "total_total_tokens_target_model": "187581.0", + "avg_total_tokens_other_models": "60424.70", + "total_total_tokens_other_models": "604247.0", + "avg_total_tokens_all_models": "79182.80", + "total_total_tokens_all_models": "791828.0", + "avg_turn_count": "31.10", + "total_turn_count": "311.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+1ep/4a160ca4 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "54349c88-e80a-4728-b181-305f97905d59,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.09240718108780466, + "uncertainty": { + "standard_deviation": 0.09240718108780466, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "10", + "stopping_reason_count_repetition_guard": "10", + "avg_total_tokens_target_model": "18758.10", + "total_total_tokens_target_model": "187581.0", + "avg_total_tokens_other_models": "60424.70", + "total_total_tokens_other_models": "604247.0", + "avg_total_tokens_all_models": "79182.80", + "total_total_tokens_all_models": "791828.0", + "avg_turn_count": "31.10", + "total_turn_count": "311.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+1ep/4a160ca4 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "54349c88-e80a-4728-b181-305f97905d59,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.30701754385964913, + "uncertainty": { + "standard_deviation": 0.09240718108780466, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "10", + "stopping_reason_count_repetition_guard": "10", + "avg_total_tokens_target_model": "18758.10", + "total_total_tokens_target_model": "187581.0", + "avg_total_tokens_other_models": "60424.70", + "total_total_tokens_other_models": "604247.0", + "avg_total_tokens_all_models": "79182.80", + "total_total_tokens_all_models": "791828.0", + "avg_turn_count": "31.10", + "total_turn_count": "311.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+1ep/4a160ca4 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "54349c88-e80a-4728-b181-305f97905d59,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": -0.02874368727071408, + "uncertainty": { + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "10", + "stopping_reason_count_repetition_guard": "10", + "avg_total_tokens_target_model": "18758.10", + "total_total_tokens_target_model": "187581.0", + "avg_total_tokens_other_models": "60424.70", + "total_total_tokens_other_models": "604247.0", + "avg_total_tokens_all_models": "79182.80", + "total_total_tokens_all_models": "791828.0", + "avg_turn_count": "31.10", + "total_turn_count": "311.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-5-20251101/caaf59c0-f77d-4878-949c-bc57d8e49cb2_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "723129c62086682b99fe1e8823d274a056e5b2cad00795e9d92cb3eaee01e534", + "total_rows": 10 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/caaf59c0-f77d-4878-949c-bc57d8e49cb2_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-5-20251101/caaf59c0-f77d-4878-949c-bc57d8e49cb2_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a3580b3d5d7d54f8abdd9235835b05cb669185ca --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/caaf59c0-f77d-4878-949c-bc57d8e49cb2_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:723129c62086682b99fe1e8823d274a056e5b2cad00795e9d92cb3eaee01e534 +size 746832 diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/d28343bb-f270-4358-9914-049cf71f9603.json b/data/healthbench/anthropic/claude-opus-4-5-20251101/d28343bb-f270-4358-9914-049cf71f9603.json new file mode 100644 index 0000000000000000000000000000000000000000..249e3c79cc57b40e29c47fa05c6f75fe3f02bc53 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/d28343bb-f270-4358-9914-049cf71f9603.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-5-20251101/1786118073.225041", + "evaluation_timestamp": "1776259833.0", + "retrieved_timestamp": "1786118073.225041", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-5-20251101", + "id": "anthropic/claude-opus-4-5-20251101", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+3ep/961799c7 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4308584502552846, + "uncertainty": { + "standard_deviation": 0.053029302269684095, + "num_samples": 90 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "16728.00", + "total_total_tokens_target_model": "200736.0", + "avg_total_tokens_other_models": "82850.00", + "total_total_tokens_other_models": "994200.0", + "avg_total_tokens_all_models": "99578.00", + "total_total_tokens_all_models": "1194936.0", + "avg_turn_count": "38.92", + "total_turn_count": "467.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+3ep/961799c7 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.053029302269684095, + "uncertainty": { + "standard_deviation": 0.053029302269684095, + "num_samples": 90 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "16728.00", + "total_total_tokens_target_model": "200736.0", + "avg_total_tokens_other_models": "82850.00", + "total_total_tokens_other_models": "994200.0", + "avg_total_tokens_all_models": "99578.00", + "total_total_tokens_all_models": "1194936.0", + "avg_turn_count": "38.92", + "total_turn_count": "467.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+3ep/961799c7 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.38468720821661995, + "uncertainty": { + "standard_deviation": 0.053029302269684095, + "num_samples": 90 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "16728.00", + "total_total_tokens_target_model": "200736.0", + "avg_total_tokens_other_models": "82850.00", + "total_total_tokens_other_models": "994200.0", + "avg_total_tokens_all_models": "99578.00", + "total_total_tokens_all_models": "1194936.0", + "avg_turn_count": "38.92", + "total_turn_count": "467.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+3ep/961799c7 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4308584502552846, + "uncertainty": { + "num_samples": 90 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "16728.00", + "total_total_tokens_target_model": "200736.0", + "avg_total_tokens_other_models": "82850.00", + "total_total_tokens_other_models": "994200.0", + "avg_total_tokens_all_models": "99578.00", + "total_total_tokens_all_models": "1194936.0", + "avg_turn_count": "38.92", + "total_turn_count": "467.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-5-20251101/d28343bb-f270-4358-9914-049cf71f9603_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "b4ba674d28e6ce2dfae6ff7e6e9dbf971dff82e14702f867ff50f8a5de16efef", + "total_rows": 90 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/d28343bb-f270-4358-9914-049cf71f9603_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-5-20251101/d28343bb-f270-4358-9914-049cf71f9603_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..804f4be68aefedbd45837babd8fb07272672329c --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/d28343bb-f270-4358-9914-049cf71f9603_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4ba674d28e6ce2dfae6ff7e6e9dbf971dff82e14702f867ff50f8a5de16efef +size 5776834 diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/e4d2982f-17a7-4436-b617-d58ccc0edc48.json b/data/healthbench/anthropic/claude-opus-4-5-20251101/e4d2982f-17a7-4436-b617-d58ccc0edc48.json new file mode 100644 index 0000000000000000000000000000000000000000..bbf1feecc2e0abca9bc4b59df6e33a7f762f089d --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/e4d2982f-17a7-4436-b617-d58ccc0edc48.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-5-20251101/1786118117.325976", + "evaluation_timestamp": "1776782181.0", + "retrieved_timestamp": "1786118117.325976", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-5-20251101", + "id": "anthropic/claude-opus-4-5-20251101", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/1e600f58 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,59c68d10-55c1-4629-b05a-bedde7f343b5" + } + }, + "evaluation_timestamp": "1776782181.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.14939983509093538, + "uncertainty": { + "standard_deviation": 0.028465067609586486, + "num_samples": 200 + }, + "details": { + "total_matched_trajectories": "50", + "stopping_reason_count_repetition_guard": "50", + "avg_total_tokens_target_model": "15652.04", + "total_total_tokens_target_model": "782602.0", + "avg_total_tokens_other_models": "63038.14", + "total_total_tokens_other_models": "3151907.0", + "avg_total_tokens_all_models": "78690.18", + "total_total_tokens_all_models": "3934509.0", + "avg_turn_count": "32.00", + "total_turn_count": "1600.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/1e600f58 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,59c68d10-55c1-4629-b05a-bedde7f343b5" + } + }, + "evaluation_timestamp": "1776782181.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.028465067609586486, + "uncertainty": { + "standard_deviation": 0.028465067609586486, + "num_samples": 200 + }, + "details": { + "total_matched_trajectories": "50", + "stopping_reason_count_repetition_guard": "50", + "avg_total_tokens_target_model": "15652.04", + "total_total_tokens_target_model": "782602.0", + "avg_total_tokens_other_models": "63038.14", + "total_total_tokens_other_models": "3151907.0", + "avg_total_tokens_all_models": "78690.18", + "total_total_tokens_all_models": "3934509.0", + "avg_turn_count": "32.00", + "total_turn_count": "1600.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/1e600f58 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,59c68d10-55c1-4629-b05a-bedde7f343b5" + } + }, + "evaluation_timestamp": "1776782181.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.22853535353535354, + "uncertainty": { + "standard_deviation": 0.028465067609586486, + "num_samples": 200 + }, + "details": { + "total_matched_trajectories": "50", + "stopping_reason_count_repetition_guard": "50", + "avg_total_tokens_target_model": "15652.04", + "total_total_tokens_target_model": "782602.0", + "avg_total_tokens_other_models": "63038.14", + "total_total_tokens_other_models": "3151907.0", + "avg_total_tokens_all_models": "78690.18", + "total_total_tokens_all_models": "3934509.0", + "avg_turn_count": "32.00", + "total_turn_count": "1600.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/1e600f58 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,59c68d10-55c1-4629-b05a-bedde7f343b5" + } + }, + "evaluation_timestamp": "1776782181.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.14939983509093538, + "uncertainty": { + "num_samples": 200 + }, + "details": { + "total_matched_trajectories": "50", + "stopping_reason_count_repetition_guard": "50", + "avg_total_tokens_target_model": "15652.04", + "total_total_tokens_target_model": "782602.0", + "avg_total_tokens_other_models": "63038.14", + "total_total_tokens_other_models": "3151907.0", + "avg_total_tokens_all_models": "78690.18", + "total_total_tokens_all_models": "3934509.0", + "avg_turn_count": "32.00", + "total_turn_count": "1600.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-5-20251101/e4d2982f-17a7-4436-b617-d58ccc0edc48_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e26872b5684db91d8a9cf516ac8e2f21ad7608a098bc214598e6318aff7a04ea", + "total_rows": 200 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-5-20251101/e4d2982f-17a7-4436-b617-d58ccc0edc48_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-5-20251101/e4d2982f-17a7-4436-b617-d58ccc0edc48_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c1b56c48527690f000053a92822531f5cbff8497 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-5-20251101/e4d2982f-17a7-4436-b617-d58ccc0edc48_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e26872b5684db91d8a9cf516ac8e2f21ad7608a098bc214598e6318aff7a04ea +size 14091553 diff --git a/data/healthbench/anthropic/claude-opus-4-6/04debb98-24fd-4a82-9e18-559fd0971a11.json b/data/healthbench/anthropic/claude-opus-4-6/04debb98-24fd-4a82-9e18-559fd0971a11.json new file mode 100644 index 0000000000000000000000000000000000000000..97e46d352dd7ddae75ec8b5185c4b42aa6e86ab1 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-6/04debb98-24fd-4a82-9e18-559fd0971a11.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-6/1786118124.14546", + "evaluation_timestamp": "1776779722.0", + "retrieved_timestamp": "1786118124.14546", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+1ep/faf7b694 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "28", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3868513489132788, + "uncertainty": { + "standard_deviation": 0.05830426661167593, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "28", + "stopping_reason_count_repetition_guard": "28", + "avg_total_tokens_target_model": "40455.50", + "total_total_tokens_target_model": "1132754.0", + "avg_total_tokens_other_models": "104317.00", + "total_total_tokens_other_models": "2920876.0", + "avg_total_tokens_all_models": "144772.50", + "total_total_tokens_all_models": "4053630.0", + "avg_turn_count": "31.75", + "total_turn_count": "889.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+1ep/faf7b694 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "28", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.05830426661167593, + "uncertainty": { + "standard_deviation": 0.05830426661167593, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "28", + "stopping_reason_count_repetition_guard": "28", + "avg_total_tokens_target_model": "40455.50", + "total_total_tokens_target_model": "1132754.0", + "avg_total_tokens_other_models": "104317.00", + "total_total_tokens_other_models": "2920876.0", + "avg_total_tokens_all_models": "144772.50", + "total_total_tokens_all_models": "4053630.0", + "avg_turn_count": "31.75", + "total_turn_count": "889.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+1ep/faf7b694 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "28", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.421875, + "uncertainty": { + "standard_deviation": 0.05830426661167593, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "28", + "stopping_reason_count_repetition_guard": "28", + "avg_total_tokens_target_model": "40455.50", + "total_total_tokens_target_model": "1132754.0", + "avg_total_tokens_other_models": "104317.00", + "total_total_tokens_other_models": "2920876.0", + "avg_total_tokens_all_models": "144772.50", + "total_total_tokens_all_models": "4053630.0", + "avg_turn_count": "31.75", + "total_turn_count": "889.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+1ep/faf7b694 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "28", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3868513489132788, + "uncertainty": { + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "28", + "stopping_reason_count_repetition_guard": "28", + "avg_total_tokens_target_model": "40455.50", + "total_total_tokens_target_model": "1132754.0", + "avg_total_tokens_other_models": "104317.00", + "total_total_tokens_other_models": "2920876.0", + "avg_total_tokens_all_models": "144772.50", + "total_total_tokens_all_models": "4053630.0", + "avg_turn_count": "31.75", + "total_turn_count": "889.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-6/04debb98-24fd-4a82-9e18-559fd0971a11_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e194f20cb7b16aea6968a1e09ca40ebea073c42bc95adc99e3119c06fbe73879", + "total_rows": 28 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-6/04debb98-24fd-4a82-9e18-559fd0971a11_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-6/04debb98-24fd-4a82-9e18-559fd0971a11_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..f56663de1751957cfd961a672481baed5f8d4c12 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-6/04debb98-24fd-4a82-9e18-559fd0971a11_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e194f20cb7b16aea6968a1e09ca40ebea073c42bc95adc99e3119c06fbe73879 +size 2584524 diff --git a/data/healthbench/anthropic/claude-opus-4-6/c687cd2c-2357-4a49-aff6-51cd4f1ff075.json b/data/healthbench/anthropic/claude-opus-4-6/c687cd2c-2357-4a49-aff6-51cd4f1ff075.json new file mode 100644 index 0000000000000000000000000000000000000000..86bd33c25385cfaba2c456e0cff450498e1c056c --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-6/c687cd2c-2357-4a49-aff6-51cd4f1ff075.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-6/1786118148.946857", + "evaluation_timestamp": "1776165242.0", + "retrieved_timestamp": "1786118148.946857", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "0b8f1d60-2081-4562-98f7-b6a976fe1c6d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5019248772414112, + "uncertainty": { + "standard_deviation": 0.0408848859633408, + "num_samples": 350 + }, + "details": { + "total_matched_trajectories": "40", + "stopping_reason_count_repetition_guard": "40", + "avg_total_tokens_target_model": "38964.85", + "total_total_tokens_target_model": "1558594.0", + "avg_total_tokens_other_models": "93816.48", + "total_total_tokens_other_models": "3752659.0", + "avg_total_tokens_all_models": "132781.33", + "total_total_tokens_all_models": "5311253.0", + "avg_turn_count": "28.38", + "total_turn_count": "1135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "0b8f1d60-2081-4562-98f7-b6a976fe1c6d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0408848859633408, + "uncertainty": { + "standard_deviation": 0.0408848859633408, + "num_samples": 350 + }, + "details": { + "total_matched_trajectories": "40", + "stopping_reason_count_repetition_guard": "40", + "avg_total_tokens_target_model": "38964.85", + "total_total_tokens_target_model": "1558594.0", + "avg_total_tokens_other_models": "93816.48", + "total_total_tokens_other_models": "3752659.0", + "avg_total_tokens_all_models": "132781.33", + "total_total_tokens_all_models": "5311253.0", + "avg_turn_count": "28.38", + "total_turn_count": "1135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "0b8f1d60-2081-4562-98f7-b6a976fe1c6d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.49016563146997927, + "uncertainty": { + "standard_deviation": 0.0408848859633408, + "num_samples": 350 + }, + "details": { + "total_matched_trajectories": "40", + "stopping_reason_count_repetition_guard": "40", + "avg_total_tokens_target_model": "38964.85", + "total_total_tokens_target_model": "1558594.0", + "avg_total_tokens_other_models": "93816.48", + "total_total_tokens_other_models": "3752659.0", + "avg_total_tokens_all_models": "132781.33", + "total_total_tokens_all_models": "5311253.0", + "avg_turn_count": "28.38", + "total_turn_count": "1135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+7ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "0b8f1d60-2081-4562-98f7-b6a976fe1c6d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5019248772414112, + "uncertainty": { + "num_samples": 350 + }, + "details": { + "total_matched_trajectories": "40", + "stopping_reason_count_repetition_guard": "40", + "avg_total_tokens_target_model": "38964.85", + "total_total_tokens_target_model": "1558594.0", + "avg_total_tokens_other_models": "93816.48", + "total_total_tokens_other_models": "3752659.0", + "avg_total_tokens_all_models": "132781.33", + "total_total_tokens_all_models": "5311253.0", + "avg_turn_count": "28.38", + "total_turn_count": "1135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-6/c687cd2c-2357-4a49-aff6-51cd4f1ff075_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "ccb3a2fb2f903d921ef34ced51b27918dfb9d772a0adfee84e9e07d46fccd5ee", + "total_rows": 350 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-6/c687cd2c-2357-4a49-aff6-51cd4f1ff075_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-6/c687cd2c-2357-4a49-aff6-51cd4f1ff075_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..eb38c4220155b1b8e873de8ce3bc815c94430823 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-6/c687cd2c-2357-4a49-aff6-51cd4f1ff075_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ccb3a2fb2f903d921ef34ced51b27918dfb9d772a0adfee84e9e07d46fccd5ee +size 24223688 diff --git a/data/healthbench/anthropic/claude-opus-4-6/e31638e7-2ee2-46e3-ad01-17e42a2a185f.json b/data/healthbench/anthropic/claude-opus-4-6/e31638e7-2ee2-46e3-ad01-17e42a2a185f.json new file mode 100644 index 0000000000000000000000000000000000000000..99522d5f98f47445623317eada76e1957c6ca010 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-6/e31638e7-2ee2-46e3-ad01-17e42a2a185f.json @@ -0,0 +1,242 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-6/1786118130.294963", + "evaluation_timestamp": "1775733996.0", + "retrieved_timestamp": "1786118130.294963", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "4", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6937510901796616, + "uncertainty": { + "standard_deviation": 0.14928482213633432, + "num_samples": 28 + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "4", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.14928482213633432, + "uncertainty": { + "standard_deviation": 0.14928482213633432, + "num_samples": 28 + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "4", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5555555555555556, + "uncertainty": { + "standard_deviation": 0.14928482213633432, + "num_samples": 28 + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+7ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "4", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6937510901796616, + "uncertainty": { + "num_samples": 28 + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-6/e31638e7-2ee2-46e3-ad01-17e42a2a185f_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "cc5c25e8bfe32fc4c874677e2f51f05918351832c66f9a1d66abc80d199393b6", + "total_rows": 28 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-6/e31638e7-2ee2-46e3-ad01-17e42a2a185f_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-6/e31638e7-2ee2-46e3-ad01-17e42a2a185f_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..64cbc7c4597080998d7df8ce2656ac8d0df9e6e6 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-6/e31638e7-2ee2-46e3-ad01-17e42a2a185f_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc5c25e8bfe32fc4c874677e2f51f05918351832c66f9a1d66abc80d199393b6 +size 2163264 diff --git a/data/healthbench/anthropic/claude-opus-4-6/fabb83aa-16f8-44c7-95f8-5448b35fdff6.json b/data/healthbench/anthropic/claude-opus-4-6/fabb83aa-16f8-44c7-95f8-5448b35fdff6.json new file mode 100644 index 0000000000000000000000000000000000000000..f747de4748b12935a57301695d90cdd22d3ab744 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-6/fabb83aa-16f8-44c7-95f8-5448b35fdff6.json @@ -0,0 +1,290 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/anthropic_claude-opus-4-6/1786118120.275655", + "evaluation_timestamp": "1776259833.0", + "retrieved_timestamp": "1786118120.275655", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "anthropic/claude-opus-4-6", + "id": "anthropic/claude-opus-4-6", + "developer": "anthropic", + "inference_platform": "anthropic", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+1ep/2ddd7ec9 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.473105971079384, + "uncertainty": { + "standard_deviation": 0.08515165210334076, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "35886.50", + "total_total_tokens_target_model": "143546.0", + "avg_total_tokens_other_models": "75462.25", + "total_total_tokens_other_models": "301849.0", + "avg_total_tokens_all_models": "111348.75", + "total_total_tokens_all_models": "445395.0", + "avg_turn_count": "31.25", + "total_turn_count": "125.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+1ep/2ddd7ec9 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.08515165210334076, + "uncertainty": { + "standard_deviation": 0.08515165210334076, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "35886.50", + "total_total_tokens_target_model": "143546.0", + "avg_total_tokens_other_models": "75462.25", + "total_total_tokens_other_models": "301849.0", + "avg_total_tokens_all_models": "111348.75", + "total_total_tokens_all_models": "445395.0", + "avg_turn_count": "31.25", + "total_turn_count": "125.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+1ep/2ddd7ec9 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4883720930232558, + "uncertainty": { + "standard_deviation": 0.08515165210334076, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "35886.50", + "total_total_tokens_target_model": "143546.0", + "avg_total_tokens_other_models": "75462.25", + "total_total_tokens_other_models": "301849.0", + "avg_total_tokens_all_models": "111348.75", + "total_total_tokens_all_models": "445395.0", + "avg_turn_count": "31.25", + "total_turn_count": "125.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+1ep/2ddd7ec9 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.473105971079384, + "uncertainty": { + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "35886.50", + "total_total_tokens_target_model": "143546.0", + "avg_total_tokens_other_models": "75462.25", + "total_total_tokens_other_models": "301849.0", + "avg_total_tokens_all_models": "111348.75", + "total_total_tokens_all_models": "445395.0", + "avg_turn_count": "31.25", + "total_turn_count": "125.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": false, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"xhigh\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/anthropic/claude-opus-4-6/fabb83aa-16f8-44c7-95f8-5448b35fdff6_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "71e47a68859b54218531480b658ff04e2cf5c4dd7f4182d19c24f3c5ea3edc90", + "total_rows": 10 + } +} \ No newline at end of file diff --git a/data/healthbench/anthropic/claude-opus-4-6/fabb83aa-16f8-44c7-95f8-5448b35fdff6_samples.jsonl b/data/healthbench/anthropic/claude-opus-4-6/fabb83aa-16f8-44c7-95f8-5448b35fdff6_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d1cfcf1cb5ed3b78affd3864dcc5016c10946c57 --- /dev/null +++ b/data/healthbench/anthropic/claude-opus-4-6/fabb83aa-16f8-44c7-95f8-5448b35fdff6_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71e47a68859b54218531480b658ff04e2cf5c4dd7f4182d19c24f3c5ea3edc90 +size 775532 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/048e84d0-b1d3-4001-a0b9-739ef7dc9b46.json b/data/healthbench/openai/gpt-5-2025-08-07/048e84d0-b1d3-4001-a0b9-739ef7dc9b46.json new file mode 100644 index 0000000000000000000000000000000000000000..d8c79a365fc91b1d1b649ecd617fe8fd7febad17 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/048e84d0-b1d3-4001-a0b9-739ef7dc9b46.json @@ -0,0 +1,254 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118005.916392", + "evaluation_timestamp": "1776259833.0", + "retrieved_timestamp": "1786118005.916392", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+1ep/5d989d99 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "6", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3724214641031299, + "uncertainty": { + "standard_deviation": 0.14143729579692474, + "num_samples": 6 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+1ep/5d989d99 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "6", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.14143729579692474, + "uncertainty": { + "standard_deviation": 0.14143729579692474, + "num_samples": 6 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+1ep/5d989d99 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "6", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.44, + "uncertainty": { + "standard_deviation": 0.14143729579692474, + "num_samples": 6 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+1ep/5d989d99 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "6", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3724214641031299, + "uncertainty": { + "num_samples": 6 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/048e84d0-b1d3-4001-a0b9-739ef7dc9b46_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "ee5987f7b8a45db3fdcbaa85946664c73ddd3f48ca5a9757e0d74b9ef4b66689", + "total_rows": 6 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/048e84d0-b1d3-4001-a0b9-739ef7dc9b46_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/048e84d0-b1d3-4001-a0b9-739ef7dc9b46_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..7baff892b23884a71f78c5b64418d38339395303 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/048e84d0-b1d3-4001-a0b9-739ef7dc9b46_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee5987f7b8a45db3fdcbaa85946664c73ddd3f48ca5a9757e0d74b9ef4b66689 +size 360962 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/04b763be-0499-4d0b-9aad-b95f94ba903e.json b/data/healthbench/openai/gpt-5-2025-08-07/04b763be-0499-4d0b-9aad-b95f94ba903e.json new file mode 100644 index 0000000000000000000000000000000000000000..c7fd92d0f7fcb1c7fcbf3c35e5363f9acfcc466e --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/04b763be-0499-4d0b-9aad-b95f94ba903e.json @@ -0,0 +1,254 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118062.119854", + "evaluation_timestamp": "1776178886.0", + "retrieved_timestamp": "1786118062.119854", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+9ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "6", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e" + } + }, + "evaluation_timestamp": "1776178886.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3935219219067352, + "uncertainty": { + "standard_deviation": 0.10812875628147743, + "num_samples": 54 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+9ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "6", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e" + } + }, + "evaluation_timestamp": "1776178886.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.10812875628147743, + "uncertainty": { + "standard_deviation": 0.10812875628147743, + "num_samples": 54 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+9ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "6", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e" + } + }, + "evaluation_timestamp": "1776178886.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5488888888888889, + "uncertainty": { + "standard_deviation": 0.10812875628147743, + "num_samples": 54 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+9ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "6", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e" + } + }, + "evaluation_timestamp": "1776178886.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3935219219067352, + "uncertainty": { + "num_samples": 54 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/04b763be-0499-4d0b-9aad-b95f94ba903e_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "fad832aa4dc943c3abd994c0ada090546bedf5be1346a54e85c5c87e11831041", + "total_rows": 54 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/04b763be-0499-4d0b-9aad-b95f94ba903e_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/04b763be-0499-4d0b-9aad-b95f94ba903e_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..4123b0ffa1129ecb8c1ea4b4d23e450cab085767 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/04b763be-0499-4d0b-9aad-b95f94ba903e_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fad832aa4dc943c3abd994c0ada090546bedf5be1346a54e85c5c87e11831041 +size 3068261 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/135c2cbd-68ee-4c3e-89eb-197eb303b289.json b/data/healthbench/openai/gpt-5-2025-08-07/135c2cbd-68ee-4c3e-89eb-197eb303b289.json new file mode 100644 index 0000000000000000000000000000000000000000..3b99cba41e32a63d7b7d1cdf2cbaa8afff35f8c6 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/135c2cbd-68ee-4c3e-89eb-197eb303b289.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118016.941975", + "evaluation_timestamp": "1776259833.0", + "retrieved_timestamp": "1786118016.941975", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep/5cf7b71f for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,1fb6ab91-a999-437d-8c93-c5937c0a89a3" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5166218009608442, + "uncertainty": { + "standard_deviation": 0.05543923991502333, + "num_samples": 60 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "25322.75", + "total_total_tokens_target_model": "202582.0", + "avg_total_tokens_other_models": "97548.25", + "total_total_tokens_other_models": "780386.0", + "avg_total_tokens_all_models": "122871.00", + "total_total_tokens_all_models": "982968.0", + "avg_turn_count": "32.50", + "total_turn_count": "260.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep/5cf7b71f for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,1fb6ab91-a999-437d-8c93-c5937c0a89a3" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.05543923991502333, + "uncertainty": { + "standard_deviation": 0.05543923991502333, + "num_samples": 60 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "25322.75", + "total_total_tokens_target_model": "202582.0", + "avg_total_tokens_other_models": "97548.25", + "total_total_tokens_other_models": "780386.0", + "avg_total_tokens_all_models": "122871.00", + "total_total_tokens_all_models": "982968.0", + "avg_turn_count": "32.50", + "total_turn_count": "260.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep/5cf7b71f for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,1fb6ab91-a999-437d-8c93-c5937c0a89a3" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.49122807017543857, + "uncertainty": { + "standard_deviation": 0.05543923991502333, + "num_samples": 60 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "25322.75", + "total_total_tokens_target_model": "202582.0", + "avg_total_tokens_other_models": "97548.25", + "total_total_tokens_other_models": "780386.0", + "avg_total_tokens_all_models": "122871.00", + "total_total_tokens_all_models": "982968.0", + "avg_turn_count": "32.50", + "total_turn_count": "260.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep/5cf7b71f for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,1fb6ab91-a999-437d-8c93-c5937c0a89a3" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5166218009608442, + "uncertainty": { + "num_samples": 60 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "25322.75", + "total_total_tokens_target_model": "202582.0", + "avg_total_tokens_other_models": "97548.25", + "total_total_tokens_other_models": "780386.0", + "avg_total_tokens_all_models": "122871.00", + "total_total_tokens_all_models": "982968.0", + "avg_turn_count": "32.50", + "total_turn_count": "260.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/135c2cbd-68ee-4c3e-89eb-197eb303b289_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "c28469e0d140ba7e94f513ae612c89a0e89c71a98aa27314e98a653a284096d5", + "total_rows": 60 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/135c2cbd-68ee-4c3e-89eb-197eb303b289_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/135c2cbd-68ee-4c3e-89eb-197eb303b289_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..52a3663c0388765115f9a3a0d0146d8f23832bcf --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/135c2cbd-68ee-4c3e-89eb-197eb303b289_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c28469e0d140ba7e94f513ae612c89a0e89c71a98aa27314e98a653a284096d5 +size 3976561 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/1bb57a90-94ca-47ec-9cb5-c697873ab2fb.json b/data/healthbench/openai/gpt-5-2025-08-07/1bb57a90-94ca-47ec-9cb5-c697873ab2fb.json new file mode 100644 index 0000000000000000000000000000000000000000..2f129353aac1099e5e377199810ce660666a6f25 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/1bb57a90-94ca-47ec-9cb5-c697873ab2fb.json @@ -0,0 +1,322 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118003.208521", + "evaluation_timestamp": "1776779722.0", + "retrieved_timestamp": "1786118003.208521", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+1ep/4f6711b0 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "53", + "sample_ids": "b160df9f-be98-4a4d-9cc8-11a612ffaf87,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,7b415a2f-22b0-4735-a38e-567f1891acc5,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.26708832639325875, + "uncertainty": { + "standard_deviation": 0.043589112369473104, + "num_samples": 52 + }, + "details": { + "total_matched_trajectories": "52", + "stopping_reason_count_repetition_guard": "50", + "stopping_reason_count_completed_with_submit": "2", + "avg_total_tokens_target_model": "23048.35", + "total_total_tokens_target_model": "1198514.0", + "avg_total_tokens_other_models": "69684.54", + "total_total_tokens_other_models": "3623596.0", + "avg_total_tokens_all_models": "92732.88", + "total_total_tokens_all_models": "4822110.0", + "avg_turn_count": "25.15", + "total_turn_count": "1308.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+1ep/4f6711b0 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "53", + "sample_ids": "b160df9f-be98-4a4d-9cc8-11a612ffaf87,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,7b415a2f-22b0-4735-a38e-567f1891acc5,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.043589112369473104, + "uncertainty": { + "standard_deviation": 0.043589112369473104, + "num_samples": 52 + }, + "details": { + "total_matched_trajectories": "52", + "stopping_reason_count_repetition_guard": "50", + "stopping_reason_count_completed_with_submit": "2", + "avg_total_tokens_target_model": "23048.35", + "total_total_tokens_target_model": "1198514.0", + "avg_total_tokens_other_models": "69684.54", + "total_total_tokens_other_models": "3623596.0", + "avg_total_tokens_all_models": "92732.88", + "total_total_tokens_all_models": "4822110.0", + "avg_turn_count": "25.15", + "total_turn_count": "1308.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+1ep/4f6711b0 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "53", + "sample_ids": "b160df9f-be98-4a4d-9cc8-11a612ffaf87,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,7b415a2f-22b0-4735-a38e-567f1891acc5,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3993055555555556, + "uncertainty": { + "standard_deviation": 0.043589112369473104, + "num_samples": 52 + }, + "details": { + "total_matched_trajectories": "52", + "stopping_reason_count_repetition_guard": "50", + "stopping_reason_count_completed_with_submit": "2", + "avg_total_tokens_target_model": "23048.35", + "total_total_tokens_target_model": "1198514.0", + "avg_total_tokens_other_models": "69684.54", + "total_total_tokens_other_models": "3623596.0", + "avg_total_tokens_all_models": "92732.88", + "total_total_tokens_all_models": "4822110.0", + "avg_turn_count": "25.15", + "total_turn_count": "1308.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+1ep/4f6711b0 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "53", + "sample_ids": "b160df9f-be98-4a4d-9cc8-11a612ffaf87,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,7b415a2f-22b0-4735-a38e-567f1891acc5,a20cbbf3-330b-46d9-a35b-92e89a022282,be6150a5-8060-4434-9724-afede307d495,3a896889-9029-4ef8-8684-ca378da4702b,a6204079-9e72-42ad-8f39-a65552fa2e68,c0d80b93-b2d5-4129-b998-b1312f98e2a8,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.26708832639325875, + "uncertainty": { + "num_samples": 52 + }, + "details": { + "total_matched_trajectories": "52", + "stopping_reason_count_repetition_guard": "50", + "stopping_reason_count_completed_with_submit": "2", + "avg_total_tokens_target_model": "23048.35", + "total_total_tokens_target_model": "1198514.0", + "avg_total_tokens_other_models": "69684.54", + "total_total_tokens_other_models": "3623596.0", + "avg_total_tokens_all_models": "92732.88", + "total_total_tokens_all_models": "4822110.0", + "avg_turn_count": "25.15", + "total_turn_count": "1308.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/1bb57a90-94ca-47ec-9cb5-c697873ab2fb_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "b0c3a9a3cde4a8a5a7b243f234e8cb4edec331e9aa093a85fb0dbe26acc8eebd", + "total_rows": 52 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/1bb57a90-94ca-47ec-9cb5-c697873ab2fb_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/1bb57a90-94ca-47ec-9cb5-c697873ab2fb_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..f885ca02a53e1249cdbb47ef617fa6b3755f2748 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/1bb57a90-94ca-47ec-9cb5-c697873ab2fb_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0c3a9a3cde4a8a5a7b243f234e8cb4edec331e9aa093a85fb0dbe26acc8eebd +size 3644847 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/2a3db812-8554-4013-aedf-dfd9fd22bd88.json b/data/healthbench/openai/gpt-5-2025-08-07/2a3db812-8554-4013-aedf-dfd9fd22bd88.json new file mode 100644 index 0000000000000000000000000000000000000000..3be8f135625213d331d36ee665817cf4cfb7caae --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/2a3db812-8554-4013-aedf-dfd9fd22bd88.json @@ -0,0 +1,254 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118041.91634", + "evaluation_timestamp": "1776272651.0", + "retrieved_timestamp": "1786118041.91634", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+8ep/731d3d60 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "c80a2a84-281d-41cc-a10b-32ee48c584f6,cf09b209-5434-4317-aa2a-7b07537b475e" + } + }, + "evaluation_timestamp": "1776272651.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.712890625, + "uncertainty": { + "standard_deviation": 0.20309817656426774, + "num_samples": 16 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+8ep/731d3d60 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "c80a2a84-281d-41cc-a10b-32ee48c584f6,cf09b209-5434-4317-aa2a-7b07537b475e" + } + }, + "evaluation_timestamp": "1776272651.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.20309817656426774, + "uncertainty": { + "standard_deviation": 0.20309817656426774, + "num_samples": 16 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+8ep/731d3d60 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "c80a2a84-281d-41cc-a10b-32ee48c584f6,cf09b209-5434-4317-aa2a-7b07537b475e" + } + }, + "evaluation_timestamp": "1776272651.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4772727272727273, + "uncertainty": { + "standard_deviation": 0.20309817656426774, + "num_samples": 16 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+8ep/731d3d60 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "2", + "sample_ids": "c80a2a84-281d-41cc-a10b-32ee48c584f6,cf09b209-5434-4317-aa2a-7b07537b475e" + } + }, + "evaluation_timestamp": "1776272651.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.712890625, + "uncertainty": { + "num_samples": 16 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/2a3db812-8554-4013-aedf-dfd9fd22bd88_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "684b66c369038af4dd603193b6d570f6059f325e76222d38031f4c2c297f8edd", + "total_rows": 16 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/2a3db812-8554-4013-aedf-dfd9fd22bd88_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/2a3db812-8554-4013-aedf-dfd9fd22bd88_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..17f24594ee9f98e667a6a44dfb7ca6037e40a4ec --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/2a3db812-8554-4013-aedf-dfd9fd22bd88_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:684b66c369038af4dd603193b6d570f6059f325e76222d38031f4c2c297f8edd +size 651437 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/3f6ddd74-ef1f-4188-9836-7d3bf43b82ba.json b/data/healthbench/openai/gpt-5-2025-08-07/3f6ddd74-ef1f-4188-9836-7d3bf43b82ba.json new file mode 100644 index 0000000000000000000000000000000000000000..1b055366e1ad5d72a9b05d828846c136bc3e29c8 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/3f6ddd74-ef1f-4188-9836-7d3bf43b82ba.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118036.46313", + "evaluation_timestamp": "1776789247.0", + "retrieved_timestamp": "1786118036.46313", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+3ep/b174319b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "1", + "sample_ids": "7074bb7a-a595-434d-883e-95ef6f5f6667" + } + }, + "evaluation_timestamp": "1776789247.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4, + "uncertainty": { + "standard_deviation": 5.551115123125783e-17, + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_repetition_guard": "3", + "avg_total_tokens_target_model": "24313.67", + "total_total_tokens_target_model": "72941.0", + "avg_total_tokens_other_models": "88996.67", + "total_total_tokens_other_models": "266990.0", + "avg_total_tokens_all_models": "113310.33", + "total_total_tokens_all_models": "339931.0", + "avg_turn_count": "31.00", + "total_turn_count": "93.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+3ep/b174319b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "1", + "sample_ids": "7074bb7a-a595-434d-883e-95ef6f5f6667" + } + }, + "evaluation_timestamp": "1776789247.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 5.551115123125783e-17, + "uncertainty": { + "standard_deviation": 5.551115123125783e-17, + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_repetition_guard": "3", + "avg_total_tokens_target_model": "24313.67", + "total_total_tokens_target_model": "72941.0", + "avg_total_tokens_other_models": "88996.67", + "total_total_tokens_other_models": "266990.0", + "avg_total_tokens_all_models": "113310.33", + "total_total_tokens_all_models": "339931.0", + "avg_turn_count": "31.00", + "total_turn_count": "93.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+3ep/b174319b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "1", + "sample_ids": "7074bb7a-a595-434d-883e-95ef6f5f6667" + } + }, + "evaluation_timestamp": "1776789247.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4523809523809524, + "uncertainty": { + "standard_deviation": 5.551115123125783e-17, + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_repetition_guard": "3", + "avg_total_tokens_target_model": "24313.67", + "total_total_tokens_target_model": "72941.0", + "avg_total_tokens_other_models": "88996.67", + "total_total_tokens_other_models": "266990.0", + "avg_total_tokens_all_models": "113310.33", + "total_total_tokens_all_models": "339931.0", + "avg_turn_count": "31.00", + "total_turn_count": "93.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+3ep/b174319b for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "1", + "sample_ids": "7074bb7a-a595-434d-883e-95ef6f5f6667" + } + }, + "evaluation_timestamp": "1776789247.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4, + "uncertainty": { + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_repetition_guard": "3", + "avg_total_tokens_target_model": "24313.67", + "total_total_tokens_target_model": "72941.0", + "avg_total_tokens_other_models": "88996.67", + "total_total_tokens_other_models": "266990.0", + "avg_total_tokens_all_models": "113310.33", + "total_total_tokens_all_models": "339931.0", + "avg_turn_count": "31.00", + "total_turn_count": "93.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/3f6ddd74-ef1f-4188-9836-7d3bf43b82ba_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "d8168ddfbf5720bff3f17d03a32bc80f6c26ce51b37429e2c3c3096a949e6d9f", + "total_rows": 3 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/3f6ddd74-ef1f-4188-9836-7d3bf43b82ba_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/3f6ddd74-ef1f-4188-9836-7d3bf43b82ba_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..8a875ab6736189357cc398efff7decf2cf55cecc --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/3f6ddd74-ef1f-4188-9836-7d3bf43b82ba_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8168ddfbf5720bff3f17d03a32bc80f6c26ce51b37429e2c3c3096a949e6d9f +size 239797 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/749b762a-6ec1-4526-a2eb-22f4df22e19e.json b/data/healthbench/openai/gpt-5-2025-08-07/749b762a-6ec1-4526-a2eb-22f4df22e19e.json new file mode 100644 index 0000000000000000000000000000000000000000..ceecc0659870ae4acbff13bbd281a06b93490cd0 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/749b762a-6ec1-4526-a2eb-22f4df22e19e.json @@ -0,0 +1,322 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118011.422175", + "evaluation_timestamp": "1776786271.0", + "retrieved_timestamp": "1786118011.422175", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep/517ba5ec for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "11", + "sample_ids": "267bda16-fd4b-4a9e-b3ac-5602c2532eed,328c55cf-0b57-4992-99b1-f6a21af62ade,a079d928-7b78-4a85-a72b-4274d7fc2914,9f7b555a-fc30-40c3-88d3-f2ece06454a2,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,69a544f8-957d-4dda-a777-608ec7b95617,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,99b22b39-30d5-40dc-82d9-c1cd3c793a44,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,36c6e780-d4a8-47b2-8a36-90ee4b97607f" + } + }, + "evaluation_timestamp": "1776786271.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3523759984042892, + "uncertainty": { + "standard_deviation": 0.04269786418416258, + "num_samples": 20 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "19", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "24947.65", + "total_total_tokens_target_model": "498953.0", + "avg_total_tokens_other_models": "71642.50", + "total_total_tokens_other_models": "1432850.0", + "avg_total_tokens_all_models": "96590.15", + "total_total_tokens_all_models": "1931803.0", + "avg_turn_count": "24.90", + "total_turn_count": "498.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep/517ba5ec for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "11", + "sample_ids": "267bda16-fd4b-4a9e-b3ac-5602c2532eed,328c55cf-0b57-4992-99b1-f6a21af62ade,a079d928-7b78-4a85-a72b-4274d7fc2914,9f7b555a-fc30-40c3-88d3-f2ece06454a2,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,69a544f8-957d-4dda-a777-608ec7b95617,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,99b22b39-30d5-40dc-82d9-c1cd3c793a44,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,36c6e780-d4a8-47b2-8a36-90ee4b97607f" + } + }, + "evaluation_timestamp": "1776786271.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.04269786418416258, + "uncertainty": { + "standard_deviation": 0.04269786418416258, + "num_samples": 20 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "19", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "24947.65", + "total_total_tokens_target_model": "498953.0", + "avg_total_tokens_other_models": "71642.50", + "total_total_tokens_other_models": "1432850.0", + "avg_total_tokens_all_models": "96590.15", + "total_total_tokens_all_models": "1931803.0", + "avg_turn_count": "24.90", + "total_turn_count": "498.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep/517ba5ec for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "11", + "sample_ids": "267bda16-fd4b-4a9e-b3ac-5602c2532eed,328c55cf-0b57-4992-99b1-f6a21af62ade,a079d928-7b78-4a85-a72b-4274d7fc2914,9f7b555a-fc30-40c3-88d3-f2ece06454a2,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,69a544f8-957d-4dda-a777-608ec7b95617,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,99b22b39-30d5-40dc-82d9-c1cd3c793a44,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,36c6e780-d4a8-47b2-8a36-90ee4b97607f" + } + }, + "evaluation_timestamp": "1776786271.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.43902439024390244, + "uncertainty": { + "standard_deviation": 0.04269786418416258, + "num_samples": 20 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "19", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "24947.65", + "total_total_tokens_target_model": "498953.0", + "avg_total_tokens_other_models": "71642.50", + "total_total_tokens_other_models": "1432850.0", + "avg_total_tokens_all_models": "96590.15", + "total_total_tokens_all_models": "1931803.0", + "avg_turn_count": "24.90", + "total_turn_count": "498.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep/517ba5ec for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "11", + "sample_ids": "267bda16-fd4b-4a9e-b3ac-5602c2532eed,328c55cf-0b57-4992-99b1-f6a21af62ade,a079d928-7b78-4a85-a72b-4274d7fc2914,9f7b555a-fc30-40c3-88d3-f2ece06454a2,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,69a544f8-957d-4dda-a777-608ec7b95617,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,99b22b39-30d5-40dc-82d9-c1cd3c793a44,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,36c6e780-d4a8-47b2-8a36-90ee4b97607f" + } + }, + "evaluation_timestamp": "1776786271.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3523759984042892, + "uncertainty": { + "num_samples": 20 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "19", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "24947.65", + "total_total_tokens_target_model": "498953.0", + "avg_total_tokens_other_models": "71642.50", + "total_total_tokens_other_models": "1432850.0", + "avg_total_tokens_all_models": "96590.15", + "total_total_tokens_all_models": "1931803.0", + "avg_turn_count": "24.90", + "total_turn_count": "498.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/749b762a-6ec1-4526-a2eb-22f4df22e19e_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "9f0ce367f99bbd49a32da341b50908cf60adf06912a649b66ea2ca6be31754d4", + "total_rows": 20 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/749b762a-6ec1-4526-a2eb-22f4df22e19e_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/749b762a-6ec1-4526-a2eb-22f4df22e19e_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..fa9b359fbd9380f317485e318fd7381a34a71b4c --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/749b762a-6ec1-4526-a2eb-22f4df22e19e_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f0ce367f99bbd49a32da341b50908cf60adf06912a649b66ea2ca6be31754d4 +size 1367604 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/8c8a80cf-c10a-4224-a3d3-4b6ff6a35bff.json b/data/healthbench/openai/gpt-5-2025-08-07/8c8a80cf-c10a-4224-a3d3-4b6ff6a35bff.json new file mode 100644 index 0000000000000000000000000000000000000000..b187fd68e5333d6ffb3d39967f8a9789d49ea31f --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/8c8a80cf-c10a-4224-a3d3-4b6ff6a35bff.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118020.124775", + "evaluation_timestamp": "1776682987.0", + "retrieved_timestamp": "1786118020.124775", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep/0ad8e915 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "8", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8" + } + }, + "evaluation_timestamp": "1776682987.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3877579124377136, + "uncertainty": { + "standard_deviation": 0.0736253118997102, + "num_samples": 15 + }, + "details": { + "total_matched_trajectories": "15", + "stopping_reason_count_repetition_guard": "15", + "avg_total_tokens_target_model": "23154.27", + "total_total_tokens_target_model": "347314.0", + "avg_total_tokens_other_models": "81239.07", + "total_total_tokens_other_models": "1218586.0", + "avg_total_tokens_all_models": "104393.33", + "total_total_tokens_all_models": "1565900.0", + "avg_turn_count": "27.40", + "total_turn_count": "411.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep/0ad8e915 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "8", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8" + } + }, + "evaluation_timestamp": "1776682987.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0736253118997102, + "uncertainty": { + "standard_deviation": 0.0736253118997102, + "num_samples": 15 + }, + "details": { + "total_matched_trajectories": "15", + "stopping_reason_count_repetition_guard": "15", + "avg_total_tokens_target_model": "23154.27", + "total_total_tokens_target_model": "347314.0", + "avg_total_tokens_other_models": "81239.07", + "total_total_tokens_other_models": "1218586.0", + "avg_total_tokens_all_models": "104393.33", + "total_total_tokens_all_models": "1565900.0", + "avg_turn_count": "27.40", + "total_turn_count": "411.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep/0ad8e915 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "8", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8" + } + }, + "evaluation_timestamp": "1776682987.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.43, + "uncertainty": { + "standard_deviation": 0.0736253118997102, + "num_samples": 15 + }, + "details": { + "total_matched_trajectories": "15", + "stopping_reason_count_repetition_guard": "15", + "avg_total_tokens_target_model": "23154.27", + "total_total_tokens_target_model": "347314.0", + "avg_total_tokens_other_models": "81239.07", + "total_total_tokens_other_models": "1218586.0", + "avg_total_tokens_all_models": "104393.33", + "total_total_tokens_all_models": "1565900.0", + "avg_turn_count": "27.40", + "total_turn_count": "411.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep/0ad8e915 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "8", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8" + } + }, + "evaluation_timestamp": "1776682987.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3877579124377136, + "uncertainty": { + "num_samples": 15 + }, + "details": { + "total_matched_trajectories": "15", + "stopping_reason_count_repetition_guard": "15", + "avg_total_tokens_target_model": "23154.27", + "total_total_tokens_target_model": "347314.0", + "avg_total_tokens_other_models": "81239.07", + "total_total_tokens_other_models": "1218586.0", + "avg_total_tokens_all_models": "104393.33", + "total_total_tokens_all_models": "1565900.0", + "avg_turn_count": "27.40", + "total_turn_count": "411.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/8c8a80cf-c10a-4224-a3d3-4b6ff6a35bff_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "03a4f9134ad33dc284950d09c88e4d1a93245e46981331a97623599d2d42c302", + "total_rows": 15 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/8c8a80cf-c10a-4224-a3d3-4b6ff6a35bff_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/8c8a80cf-c10a-4224-a3d3-4b6ff6a35bff_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..5a66fe1d56be342476b670f2ca06a2b024eb4fc4 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/8c8a80cf-c10a-4224-a3d3-4b6ff6a35bff_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03a4f9134ad33dc284950d09c88e4d1a93245e46981331a97623599d2d42c302 +size 1104593 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/cf03a584-f431-440f-b091-369c89a7916c.json b/data/healthbench/openai/gpt-5-2025-08-07/cf03a584-f431-440f-b091-369c89a7916c.json new file mode 100644 index 0000000000000000000000000000000000000000..d0819d7733e21fbaa5342bfb2721d00c8141cfac --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/cf03a584-f431-440f-b091-369c89a7916c.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118039.570612", + "evaluation_timestamp": "1776335705.0", + "retrieved_timestamp": "1786118039.570612", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+7ep/88f76f77 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "4", + "sample_ids": "48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5289580454823439, + "uncertainty": { + "standard_deviation": 0.15480172722574595, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_repetition_guard": "5", + "avg_total_tokens_target_model": "26514.40", + "total_total_tokens_target_model": "132572.0", + "avg_total_tokens_other_models": "37801.00", + "total_total_tokens_other_models": "189005.0", + "avg_total_tokens_all_models": "64315.40", + "total_total_tokens_all_models": "321577.0", + "avg_turn_count": "15.00", + "total_turn_count": "75.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+7ep/88f76f77 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "4", + "sample_ids": "48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.15480172722574595, + "uncertainty": { + "standard_deviation": 0.15480172722574595, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_repetition_guard": "5", + "avg_total_tokens_target_model": "26514.40", + "total_total_tokens_target_model": "132572.0", + "avg_total_tokens_other_models": "37801.00", + "total_total_tokens_other_models": "189005.0", + "avg_total_tokens_all_models": "64315.40", + "total_total_tokens_all_models": "321577.0", + "avg_turn_count": "15.00", + "total_turn_count": "75.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+7ep/88f76f77 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "4", + "sample_ids": "48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.49261083743842365, + "uncertainty": { + "standard_deviation": 0.15480172722574595, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_repetition_guard": "5", + "avg_total_tokens_target_model": "26514.40", + "total_total_tokens_target_model": "132572.0", + "avg_total_tokens_other_models": "37801.00", + "total_total_tokens_other_models": "189005.0", + "avg_total_tokens_all_models": "64315.40", + "total_total_tokens_all_models": "321577.0", + "avg_turn_count": "15.00", + "total_turn_count": "75.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+7ep/88f76f77 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "4", + "sample_ids": "48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776335705.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5289580454823439, + "uncertainty": { + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_repetition_guard": "5", + "avg_total_tokens_target_model": "26514.40", + "total_total_tokens_target_model": "132572.0", + "avg_total_tokens_other_models": "37801.00", + "total_total_tokens_other_models": "189005.0", + "avg_total_tokens_all_models": "64315.40", + "total_total_tokens_all_models": "321577.0", + "avg_turn_count": "15.00", + "total_turn_count": "75.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/cf03a584-f431-440f-b091-369c89a7916c_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "d6f41ef0e07bcb7d4dcd811f8cb131b5666f17907dd00d60e79f3fad8d1eea6c", + "total_rows": 28 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/cf03a584-f431-440f-b091-369c89a7916c_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/cf03a584-f431-440f-b091-369c89a7916c_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a43cbe5edca396e53064b49aa4281db5921f297d --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/cf03a584-f431-440f-b091-369c89a7916c_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6f41ef0e07bcb7d4dcd811f8cb131b5666f17907dd00d60e79f3fad8d1eea6c +size 2167859 diff --git a/data/healthbench/openai/gpt-5-2025-08-07/e0cf2a83-abff-4621-8008-0a6c4baf7cc8.json b/data/healthbench/openai/gpt-5-2025-08-07/e0cf2a83-abff-4621-8008-0a6c4baf7cc8.json new file mode 100644 index 0000000000000000000000000000000000000000..f349d3af2908bb2c25e2b2a2983f508d8a861612 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/e0cf2a83-abff-4621-8008-0a6c4baf7cc8.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5-2025-08-07/1786118057.342181", + "evaluation_timestamp": "1776165242.0", + "retrieved_timestamp": "1786118057.342181", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5-2025-08-07", + "id": "openai/gpt-5-2025-08-07", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,1fb6ab91-a999-437d-8c93-c5937c0a89a3" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5344810385401033, + "uncertainty": { + "standard_deviation": 0.05002981069974894, + "num_samples": 240 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "24943.00", + "total_total_tokens_target_model": "299316.0", + "avg_total_tokens_other_models": "94987.50", + "total_total_tokens_other_models": "1139850.0", + "avg_total_tokens_all_models": "119930.50", + "total_total_tokens_all_models": "1439166.0", + "avg_turn_count": "32.50", + "total_turn_count": "390.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,1fb6ab91-a999-437d-8c93-c5937c0a89a3" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.05002981069974894, + "uncertainty": { + "standard_deviation": 0.05002981069974894, + "num_samples": 240 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "24943.00", + "total_total_tokens_target_model": "299316.0", + "avg_total_tokens_other_models": "94987.50", + "total_total_tokens_other_models": "1139850.0", + "avg_total_tokens_all_models": "119930.50", + "total_total_tokens_all_models": "1439166.0", + "avg_turn_count": "32.50", + "total_turn_count": "390.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,1fb6ab91-a999-437d-8c93-c5937c0a89a3" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5109649122807017, + "uncertainty": { + "standard_deviation": 0.05002981069974894, + "num_samples": 240 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "24943.00", + "total_total_tokens_target_model": "299316.0", + "avg_total_tokens_other_models": "94987.50", + "total_total_tokens_other_models": "1139850.0", + "avg_total_tokens_all_models": "119930.50", + "total_total_tokens_all_models": "1439166.0", + "avg_turn_count": "32.50", + "total_turn_count": "390.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+8ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,1fb6ab91-a999-437d-8c93-c5937c0a89a3" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5344810385401033, + "uncertainty": { + "num_samples": 240 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "24943.00", + "total_total_tokens_target_model": "299316.0", + "avg_total_tokens_other_models": "94987.50", + "total_total_tokens_other_models": "1139850.0", + "avg_total_tokens_all_models": "119930.50", + "total_total_tokens_all_models": "1439166.0", + "avg_turn_count": "32.50", + "total_turn_count": "390.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5-2025-08-07/e0cf2a83-abff-4621-8008-0a6c4baf7cc8_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "d8cc16fb8df75985ae4898164c80ccd554b647f9a21f8f4fe54378bb2b9b3764", + "total_rows": 240 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5-2025-08-07/e0cf2a83-abff-4621-8008-0a6c4baf7cc8_samples.jsonl b/data/healthbench/openai/gpt-5-2025-08-07/e0cf2a83-abff-4621-8008-0a6c4baf7cc8_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a520fd5e450cf527982b6565798d9e36ccfe34e7 --- /dev/null +++ b/data/healthbench/openai/gpt-5-2025-08-07/e0cf2a83-abff-4621-8008-0a6c4baf7cc8_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8cc16fb8df75985ae4898164c80ccd554b647f9a21f8f4fe54378bb2b9b3764 +size 15942839 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/0013c3d6-41f5-4867-8f98-e9043f2d5d55.json b/data/healthbench/openai/gpt-5.2-2025-12-11/0013c3d6-41f5-4867-8f98-e9043f2d5d55.json new file mode 100644 index 0000000000000000000000000000000000000000..4531a9bcff9a7f32e1f46e2cb261966141a5548f --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/0013c3d6-41f5-4867-8f98-e9043f2d5d55.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117815.355406", + "evaluation_timestamp": "1776107988.0", + "retrieved_timestamp": "1786117815.355406", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "12", + "sample_ids": "bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2" + } + }, + "evaluation_timestamp": "1776107988.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.614861651515878, + "uncertainty": { + "standard_deviation": 0.09350775326713912, + "num_samples": 24 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "6", + "avg_total_tokens_target_model": "13563.17", + "total_total_tokens_target_model": "81379.0", + "avg_total_tokens_other_models": "48012.50", + "total_total_tokens_other_models": "288075.0", + "avg_total_tokens_all_models": "61575.67", + "total_total_tokens_all_models": "369454.0", + "avg_turn_count": "20.33", + "total_turn_count": "122.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "12", + "sample_ids": "bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2" + } + }, + "evaluation_timestamp": "1776107988.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.09350775326713912, + "uncertainty": { + "standard_deviation": 0.09350775326713912, + "num_samples": 24 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "6", + "avg_total_tokens_target_model": "13563.17", + "total_total_tokens_target_model": "81379.0", + "avg_total_tokens_other_models": "48012.50", + "total_total_tokens_other_models": "288075.0", + "avg_total_tokens_all_models": "61575.67", + "total_total_tokens_all_models": "369454.0", + "avg_turn_count": "20.33", + "total_turn_count": "122.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "12", + "sample_ids": "bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2" + } + }, + "evaluation_timestamp": "1776107988.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.49606299212598426, + "uncertainty": { + "standard_deviation": 0.09350775326713912, + "num_samples": 24 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "6", + "avg_total_tokens_target_model": "13563.17", + "total_total_tokens_target_model": "81379.0", + "avg_total_tokens_other_models": "48012.50", + "total_total_tokens_other_models": "288075.0", + "avg_total_tokens_all_models": "61575.67", + "total_total_tokens_all_models": "369454.0", + "avg_turn_count": "20.33", + "total_turn_count": "122.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "12", + "sample_ids": "bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2" + } + }, + "evaluation_timestamp": "1776107988.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.614861651515878, + "uncertainty": { + "num_samples": 24 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "6", + "avg_total_tokens_target_model": "13563.17", + "total_total_tokens_target_model": "81379.0", + "avg_total_tokens_other_models": "48012.50", + "total_total_tokens_other_models": "288075.0", + "avg_total_tokens_all_models": "61575.67", + "total_total_tokens_all_models": "369454.0", + "avg_turn_count": "20.33", + "total_turn_count": "122.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/0013c3d6-41f5-4867-8f98-e9043f2d5d55_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "81a09f4aae17329678429e011f9349c0bb1d52f2b3127114b425af1ef8a08fe6", + "total_rows": 24 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/0013c3d6-41f5-4867-8f98-e9043f2d5d55_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/0013c3d6-41f5-4867-8f98-e9043f2d5d55_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b706e5f6ce9b71931a979f91157b475f84899a28 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/0013c3d6-41f5-4867-8f98-e9043f2d5d55_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:81a09f4aae17329678429e011f9349c0bb1d52f2b3127114b425af1ef8a08fe6 +size 1444496 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/0e7de2c9-84ac-40d4-9716-56efb38dadd2.json b/data/healthbench/openai/gpt-5.2-2025-12-11/0e7de2c9-84ac-40d4-9716-56efb38dadd2.json new file mode 100644 index 0000000000000000000000000000000000000000..cf8b2ab3e41e04312cafb26a0736b8f837b1caa4 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/0e7de2c9-84ac-40d4-9716-56efb38dadd2.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117827.085429", + "evaluation_timestamp": "1776259833.0", + "retrieved_timestamp": "1786117827.085429", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep/51835bb6 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "53", + "sample_ids": "bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.489585617148727, + "uncertainty": { + "standard_deviation": 0.034273865305307856, + "num_samples": 106 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "13346.50", + "total_total_tokens_target_model": "266930.0", + "avg_total_tokens_other_models": "59882.70", + "total_total_tokens_other_models": "1197654.0", + "avg_total_tokens_all_models": "73229.20", + "total_total_tokens_all_models": "1464584.0", + "avg_turn_count": "27.80", + "total_turn_count": "556.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep/51835bb6 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "53", + "sample_ids": "bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.034273865305307856, + "uncertainty": { + "standard_deviation": 0.034273865305307856, + "num_samples": 106 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "13346.50", + "total_total_tokens_target_model": "266930.0", + "avg_total_tokens_other_models": "59882.70", + "total_total_tokens_other_models": "1197654.0", + "avg_total_tokens_all_models": "73229.20", + "total_total_tokens_all_models": "1464584.0", + "avg_turn_count": "27.80", + "total_turn_count": "556.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep/51835bb6 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "53", + "sample_ids": "bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.46195652173913043, + "uncertainty": { + "standard_deviation": 0.034273865305307856, + "num_samples": 106 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "13346.50", + "total_total_tokens_target_model": "266930.0", + "avg_total_tokens_other_models": "59882.70", + "total_total_tokens_other_models": "1197654.0", + "avg_total_tokens_all_models": "73229.20", + "total_total_tokens_all_models": "1464584.0", + "avg_turn_count": "27.80", + "total_turn_count": "556.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep/51835bb6 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "53", + "sample_ids": "bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed,c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.489585617148727, + "uncertainty": { + "num_samples": 106 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "13346.50", + "total_total_tokens_target_model": "266930.0", + "avg_total_tokens_other_models": "59882.70", + "total_total_tokens_other_models": "1197654.0", + "avg_total_tokens_all_models": "73229.20", + "total_total_tokens_all_models": "1464584.0", + "avg_turn_count": "27.80", + "total_turn_count": "556.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/0e7de2c9-84ac-40d4-9716-56efb38dadd2_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "ade31863a28d06fc2e9a9ae47458ff5dec2a0393302936b669f81b61e0ee6bf8", + "total_rows": 106 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/0e7de2c9-84ac-40d4-9716-56efb38dadd2_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/0e7de2c9-84ac-40d4-9716-56efb38dadd2_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..af0b082147d55c3b43d1d02f94556caa83c29985 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/0e7de2c9-84ac-40d4-9716-56efb38dadd2_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ade31863a28d06fc2e9a9ae47458ff5dec2a0393302936b669f81b61e0ee6bf8 +size 6989368 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/0eef161a-2da9-4410-a4b1-aeca09798e9e.json b/data/healthbench/openai/gpt-5.2-2025-12-11/0eef161a-2da9-4410-a4b1-aeca09798e9e.json new file mode 100644 index 0000000000000000000000000000000000000000..ffb995de231043a0145199f2d98d13210b2d4e11 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/0eef161a-2da9-4410-a4b1-aeca09798e9e.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117834.544358", + "evaluation_timestamp": "1776696116.0", + "retrieved_timestamp": "1786117834.544358", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+3ep/8d43239b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "20", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3857739857663652, + "uncertainty": { + "standard_deviation": 0.05751152441255183, + "num_samples": 59 + }, + "details": { + "total_matched_trajectories": "59", + "stopping_reason_count_repetition_guard": "59", + "avg_total_tokens_target_model": "11293.32", + "total_total_tokens_target_model": "666306.0", + "avg_total_tokens_other_models": "57151.17", + "total_total_tokens_other_models": "3371919.0", + "avg_total_tokens_all_models": "68444.49", + "total_total_tokens_all_models": "4038225.0", + "avg_turn_count": "25.46", + "total_turn_count": "1502.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+3ep/8d43239b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "20", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.05751152441255183, + "uncertainty": { + "standard_deviation": 0.05751152441255183, + "num_samples": 59 + }, + "details": { + "total_matched_trajectories": "59", + "stopping_reason_count_repetition_guard": "59", + "avg_total_tokens_target_model": "11293.32", + "total_total_tokens_target_model": "666306.0", + "avg_total_tokens_other_models": "57151.17", + "total_total_tokens_other_models": "3371919.0", + "avg_total_tokens_all_models": "68444.49", + "total_total_tokens_all_models": "4038225.0", + "avg_turn_count": "25.46", + "total_turn_count": "1502.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+3ep/8d43239b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "20", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4147982062780269, + "uncertainty": { + "standard_deviation": 0.05751152441255183, + "num_samples": 59 + }, + "details": { + "total_matched_trajectories": "59", + "stopping_reason_count_repetition_guard": "59", + "avg_total_tokens_target_model": "11293.32", + "total_total_tokens_target_model": "666306.0", + "avg_total_tokens_other_models": "57151.17", + "total_total_tokens_other_models": "3371919.0", + "avg_total_tokens_all_models": "68444.49", + "total_total_tokens_all_models": "4038225.0", + "avg_turn_count": "25.46", + "total_turn_count": "1502.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+3ep/8d43239b for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "20", + "sample_ids": "c3eeecfc-de96-480f-929d-ac4a5e5a541c,269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade" + } + }, + "evaluation_timestamp": "1776696116.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3857739857663652, + "uncertainty": { + "num_samples": 59 + }, + "details": { + "total_matched_trajectories": "59", + "stopping_reason_count_repetition_guard": "59", + "avg_total_tokens_target_model": "11293.32", + "total_total_tokens_target_model": "666306.0", + "avg_total_tokens_other_models": "57151.17", + "total_total_tokens_other_models": "3371919.0", + "avg_total_tokens_all_models": "68444.49", + "total_total_tokens_all_models": "4038225.0", + "avg_turn_count": "25.46", + "total_turn_count": "1502.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/0eef161a-2da9-4410-a4b1-aeca09798e9e_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "aec67e0f4157e243735801637f21679481ac5cb78bf4a4bdaf068d745efdbce9", + "total_rows": 59 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/0eef161a-2da9-4410-a4b1-aeca09798e9e_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/0eef161a-2da9-4410-a4b1-aeca09798e9e_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..f900917e586d322b9ef5736387f4af11e904bc26 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/0eef161a-2da9-4410-a4b1-aeca09798e9e_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aec67e0f4157e243735801637f21679481ac5cb78bf4a4bdaf068d745efdbce9 +size 3857028 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/1949aa2e-9bac-469a-985d-0aa6bd5c82d5.json b/data/healthbench/openai/gpt-5.2-2025-12-11/1949aa2e-9bac-469a-985d-0aa6bd5c82d5.json new file mode 100644 index 0000000000000000000000000000000000000000..e465332cf08309f7dc5f3b25c620594690e1b775 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/1949aa2e-9bac-469a-985d-0aa6bd5c82d5.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117886.693749", + "evaluation_timestamp": "1776789503.0", + "retrieved_timestamp": "1786117886.693749", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/7aaca4a0 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "1", + "sample_ids": "c835bebe-e998-4dcd-843a-2de03d4aabc6" + } + }, + "evaluation_timestamp": "1776789503.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.34090909090909094, + "uncertainty": { + "standard_deviation": 5.551115123125783e-17, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "16733.00", + "total_total_tokens_target_model": "16733.0", + "avg_total_tokens_other_models": "91142.00", + "total_total_tokens_other_models": "91142.0", + "avg_total_tokens_all_models": "107875.00", + "total_total_tokens_all_models": "107875.0", + "avg_turn_count": "36.00", + "total_turn_count": "36.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/7aaca4a0 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "1", + "sample_ids": "c835bebe-e998-4dcd-843a-2de03d4aabc6" + } + }, + "evaluation_timestamp": "1776789503.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 5.551115123125783e-17, + "uncertainty": { + "standard_deviation": 5.551115123125783e-17, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "16733.00", + "total_total_tokens_target_model": "16733.0", + "avg_total_tokens_other_models": "91142.00", + "total_total_tokens_other_models": "91142.0", + "avg_total_tokens_all_models": "107875.00", + "total_total_tokens_all_models": "107875.0", + "avg_turn_count": "36.00", + "total_turn_count": "36.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/7aaca4a0 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "1", + "sample_ids": "c835bebe-e998-4dcd-843a-2de03d4aabc6" + } + }, + "evaluation_timestamp": "1776789503.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.40625, + "uncertainty": { + "standard_deviation": 5.551115123125783e-17, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "16733.00", + "total_total_tokens_target_model": "16733.0", + "avg_total_tokens_other_models": "91142.00", + "total_total_tokens_other_models": "91142.0", + "avg_total_tokens_all_models": "107875.00", + "total_total_tokens_all_models": "107875.0", + "avg_turn_count": "36.00", + "total_turn_count": "36.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/7aaca4a0 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "1", + "sample_ids": "c835bebe-e998-4dcd-843a-2de03d4aabc6" + } + }, + "evaluation_timestamp": "1776789503.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.34090909090909094, + "uncertainty": { + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "16733.00", + "total_total_tokens_target_model": "16733.0", + "avg_total_tokens_other_models": "91142.00", + "total_total_tokens_other_models": "91142.0", + "avg_total_tokens_all_models": "107875.00", + "total_total_tokens_all_models": "107875.0", + "avg_turn_count": "36.00", + "total_turn_count": "36.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/1949aa2e-9bac-469a-985d-0aa6bd5c82d5_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "f42d5c12712ad308d49a2279ab61f2098389fbf9de44a413e5de62cbc75a70aa", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/1949aa2e-9bac-469a-985d-0aa6bd5c82d5_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/1949aa2e-9bac-469a-985d-0aa6bd5c82d5_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b367e8338f2b6fd5b743781c290cbb942bad3871 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/1949aa2e-9bac-469a-985d-0aa6bd5c82d5_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f42d5c12712ad308d49a2279ab61f2098389fbf9de44a413e5de62cbc75a70aa +size 357538 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/1b76356b-e91a-4f79-b9be-facad8fdf208.json b/data/healthbench/openai/gpt-5.2-2025-12-11/1b76356b-e91a-4f79-b9be-facad8fdf208.json new file mode 100644 index 0000000000000000000000000000000000000000..8a86ed329270c4cf5a7302172316e346b48494f6 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/1b76356b-e91a-4f79-b9be-facad8fdf208.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117880.651714", + "evaluation_timestamp": "1776165242.0", + "retrieved_timestamp": "1786117880.651714", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "7", + "sample_ids": "6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5469407587970319, + "uncertainty": { + "standard_deviation": 0.08930226417991032, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "6", + "avg_total_tokens_target_model": "13051.50", + "total_total_tokens_target_model": "78309.0", + "avg_total_tokens_other_models": "70890.17", + "total_total_tokens_other_models": "425341.0", + "avg_total_tokens_all_models": "83941.67", + "total_total_tokens_all_models": "503650.0", + "avg_turn_count": "30.33", + "total_turn_count": "182.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "7", + "sample_ids": "6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.08930226417991032, + "uncertainty": { + "standard_deviation": 0.08930226417991032, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "6", + "avg_total_tokens_target_model": "13051.50", + "total_total_tokens_target_model": "78309.0", + "avg_total_tokens_other_models": "70890.17", + "total_total_tokens_other_models": "425341.0", + "avg_total_tokens_all_models": "83941.67", + "total_total_tokens_all_models": "503650.0", + "avg_turn_count": "30.33", + "total_turn_count": "182.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "7", + "sample_ids": "6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4976851851851852, + "uncertainty": { + "standard_deviation": 0.08930226417991032, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "6", + "avg_total_tokens_target_model": "13051.50", + "total_total_tokens_target_model": "78309.0", + "avg_total_tokens_other_models": "70890.17", + "total_total_tokens_other_models": "425341.0", + "avg_total_tokens_all_models": "83941.67", + "total_total_tokens_all_models": "503650.0", + "avg_turn_count": "30.33", + "total_turn_count": "182.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "7", + "sample_ids": "6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5469407587970319, + "uncertainty": { + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "6", + "avg_total_tokens_target_model": "13051.50", + "total_total_tokens_target_model": "78309.0", + "avg_total_tokens_other_models": "70890.17", + "total_total_tokens_other_models": "425341.0", + "avg_total_tokens_all_models": "83941.67", + "total_total_tokens_all_models": "503650.0", + "avg_turn_count": "30.33", + "total_turn_count": "182.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/1b76356b-e91a-4f79-b9be-facad8fdf208_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "d88f217f8c06a225c5689bc2965530d1e0aa38281354f946fcc53ccbe35cde8e", + "total_rows": 28 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/1b76356b-e91a-4f79-b9be-facad8fdf208_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/1b76356b-e91a-4f79-b9be-facad8fdf208_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..931402c469a14b44b7776ee414f987818d03195b --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/1b76356b-e91a-4f79-b9be-facad8fdf208_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d88f217f8c06a225c5689bc2965530d1e0aa38281354f946fcc53ccbe35cde8e +size 2262815 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/3bded993-8406-4f5a-9ea5-9d7605ccdd20.json b/data/healthbench/openai/gpt-5.2-2025-12-11/3bded993-8406-4f5a-9ea5-9d7605ccdd20.json new file mode 100644 index 0000000000000000000000000000000000000000..b5a3ee4157709aaa23050a79bc0343451e257c5e --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/3bded993-8406-4f5a-9ea5-9d7605ccdd20.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117812.661763", + "evaluation_timestamp": "1776779722.0", + "retrieved_timestamp": "1786117812.661763", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+1ep/53b419e3 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "28", + "sample_ids": "be6150a5-8060-4434-9724-afede307d495,7074bb7a-a595-434d-883e-95ef6f5f6667,a6204079-9e72-42ad-8f39-a65552fa2e68,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.287102978548937, + "uncertainty": { + "standard_deviation": 0.04943948877896748, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "28", + "stopping_reason_count_repetition_guard": "28", + "avg_total_tokens_target_model": "13433.79", + "total_total_tokens_target_model": "376146.0", + "avg_total_tokens_other_models": "64395.00", + "total_total_tokens_other_models": "1803060.0", + "avg_total_tokens_all_models": "77828.79", + "total_total_tokens_all_models": "2179206.0", + "avg_turn_count": "26.61", + "total_turn_count": "745.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+1ep/53b419e3 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "28", + "sample_ids": "be6150a5-8060-4434-9724-afede307d495,7074bb7a-a595-434d-883e-95ef6f5f6667,a6204079-9e72-42ad-8f39-a65552fa2e68,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.04943948877896748, + "uncertainty": { + "standard_deviation": 0.04943948877896748, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "28", + "stopping_reason_count_repetition_guard": "28", + "avg_total_tokens_target_model": "13433.79", + "total_total_tokens_target_model": "376146.0", + "avg_total_tokens_other_models": "64395.00", + "total_total_tokens_other_models": "1803060.0", + "avg_total_tokens_all_models": "77828.79", + "total_total_tokens_all_models": "2179206.0", + "avg_turn_count": "26.61", + "total_turn_count": "745.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+1ep/53b419e3 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "28", + "sample_ids": "be6150a5-8060-4434-9724-afede307d495,7074bb7a-a595-434d-883e-95ef6f5f6667,a6204079-9e72-42ad-8f39-a65552fa2e68,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3761467889908257, + "uncertainty": { + "standard_deviation": 0.04943948877896748, + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "28", + "stopping_reason_count_repetition_guard": "28", + "avg_total_tokens_target_model": "13433.79", + "total_total_tokens_target_model": "376146.0", + "avg_total_tokens_other_models": "64395.00", + "total_total_tokens_other_models": "1803060.0", + "avg_total_tokens_all_models": "77828.79", + "total_total_tokens_all_models": "2179206.0", + "avg_turn_count": "26.61", + "total_turn_count": "745.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+1ep/53b419e3 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "28", + "sample_ids": "be6150a5-8060-4434-9724-afede307d495,7074bb7a-a595-434d-883e-95ef6f5f6667,a6204079-9e72-42ad-8f39-a65552fa2e68,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.287102978548937, + "uncertainty": { + "num_samples": 28 + }, + "details": { + "total_matched_trajectories": "28", + "stopping_reason_count_repetition_guard": "28", + "avg_total_tokens_target_model": "13433.79", + "total_total_tokens_target_model": "376146.0", + "avg_total_tokens_other_models": "64395.00", + "total_total_tokens_other_models": "1803060.0", + "avg_total_tokens_all_models": "77828.79", + "total_total_tokens_all_models": "2179206.0", + "avg_turn_count": "26.61", + "total_turn_count": "745.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/3bded993-8406-4f5a-9ea5-9d7605ccdd20_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "65e7d5bb38031eebfbe6429ab189c7607b6899b37082dcf30012c3c98b1b45d4", + "total_rows": 28 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/3bded993-8406-4f5a-9ea5-9d7605ccdd20_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/3bded993-8406-4f5a-9ea5-9d7605ccdd20_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a00a69201a668f7c2525d549999efc3494c0738d --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/3bded993-8406-4f5a-9ea5-9d7605ccdd20_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65e7d5bb38031eebfbe6429ab189c7607b6899b37082dcf30012c3c98b1b45d4 +size 1947769 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/874fb122-0288-4b98-904b-54da0931ae3c.json b/data/healthbench/openai/gpt-5.2-2025-12-11/874fb122-0288-4b98-904b-54da0931ae3c.json new file mode 100644 index 0000000000000000000000000000000000000000..e0856b0653a882ca7679289a90fe1ffb33c2ce9d --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/874fb122-0288-4b98-904b-54da0931ae3c.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117905.04423", + "evaluation_timestamp": "1775738242.0", + "retrieved_timestamp": "1786117905.04423", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6" + } + }, + "evaluation_timestamp": "1775738242.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.501916935927166, + "uncertainty": { + "standard_deviation": 0.046426074278503406, + "num_samples": 259 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "12554.50", + "total_total_tokens_target_model": "100436.0", + "avg_total_tokens_other_models": "72033.88", + "total_total_tokens_other_models": "576271.0", + "avg_total_tokens_all_models": "84588.38", + "total_total_tokens_all_models": "676707.0", + "avg_turn_count": "32.75", + "total_turn_count": "262.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6" + } + }, + "evaluation_timestamp": "1775738242.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.046426074278503406, + "uncertainty": { + "standard_deviation": 0.046426074278503406, + "num_samples": 259 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "12554.50", + "total_total_tokens_target_model": "100436.0", + "avg_total_tokens_other_models": "72033.88", + "total_total_tokens_other_models": "576271.0", + "avg_total_tokens_all_models": "84588.38", + "total_total_tokens_all_models": "676707.0", + "avg_turn_count": "32.75", + "total_turn_count": "262.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6" + } + }, + "evaluation_timestamp": "1775738242.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4848592803705022, + "uncertainty": { + "standard_deviation": 0.046426074278503406, + "num_samples": 259 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "12554.50", + "total_total_tokens_target_model": "100436.0", + "avg_total_tokens_other_models": "72033.88", + "total_total_tokens_other_models": "576271.0", + "avg_total_tokens_all_models": "84588.38", + "total_total_tokens_all_models": "676707.0", + "avg_turn_count": "32.75", + "total_turn_count": "262.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+7ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6" + } + }, + "evaluation_timestamp": "1775738242.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.501916935927166, + "uncertainty": { + "num_samples": 259 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "12554.50", + "total_total_tokens_target_model": "100436.0", + "avg_total_tokens_other_models": "72033.88", + "total_total_tokens_other_models": "576271.0", + "avg_total_tokens_all_models": "84588.38", + "total_total_tokens_all_models": "676707.0", + "avg_turn_count": "32.75", + "total_turn_count": "262.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/874fb122-0288-4b98-904b-54da0931ae3c_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e4babc593a49eea485042b9838dedd3ed2fdc77cd095cac4da8c52b1a0ff93a7", + "total_rows": 259 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/874fb122-0288-4b98-904b-54da0931ae3c_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/874fb122-0288-4b98-904b-54da0931ae3c_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..2826c544041eecf8ff79f5c83c485834c342d7b7 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/874fb122-0288-4b98-904b-54da0931ae3c_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4babc593a49eea485042b9838dedd3ed2fdc77cd095cac4da8c52b1a0ff93a7 +size 15516237 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/89899de3-59ad-41c3-8b09-63f587910abe.json b/data/healthbench/openai/gpt-5.2-2025-12-11/89899de3-59ad-41c3-8b09-63f587910abe.json new file mode 100644 index 0000000000000000000000000000000000000000..37bec02daaf33b9dcdbf16faf80ffa62052de6b0 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/89899de3-59ad-41c3-8b09-63f587910abe.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117884.588039", + "evaluation_timestamp": "1776784808.0", + "retrieved_timestamp": "1786117884.588039", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/36a174ea for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "9", + "sample_ids": "20cdebb7-9189-4f09-ad21-bfc8e8b75680,6f827e79-05c4-4383-8ed4-16d36260f29f,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776784808.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.45674974470724616, + "uncertainty": { + "standard_deviation": 0.050229918828339014, + "num_samples": 36 + }, + "details": { + "total_matched_trajectories": "9", + "stopping_reason_count_repetition_guard": "9", + "avg_total_tokens_target_model": "11154.00", + "total_total_tokens_target_model": "100386.0", + "avg_total_tokens_other_models": "57830.22", + "total_total_tokens_other_models": "520472.0", + "avg_total_tokens_all_models": "68984.22", + "total_total_tokens_all_models": "620858.0", + "avg_turn_count": "26.78", + "total_turn_count": "241.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/36a174ea for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "9", + "sample_ids": "20cdebb7-9189-4f09-ad21-bfc8e8b75680,6f827e79-05c4-4383-8ed4-16d36260f29f,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776784808.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.050229918828339014, + "uncertainty": { + "standard_deviation": 0.050229918828339014, + "num_samples": 36 + }, + "details": { + "total_matched_trajectories": "9", + "stopping_reason_count_repetition_guard": "9", + "avg_total_tokens_target_model": "11154.00", + "total_total_tokens_target_model": "100386.0", + "avg_total_tokens_other_models": "57830.22", + "total_total_tokens_other_models": "520472.0", + "avg_total_tokens_all_models": "68984.22", + "total_total_tokens_all_models": "620858.0", + "avg_turn_count": "26.78", + "total_turn_count": "241.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/36a174ea for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "9", + "sample_ids": "20cdebb7-9189-4f09-ad21-bfc8e8b75680,6f827e79-05c4-4383-8ed4-16d36260f29f,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776784808.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4716981132075472, + "uncertainty": { + "standard_deviation": 0.050229918828339014, + "num_samples": 36 + }, + "details": { + "total_matched_trajectories": "9", + "stopping_reason_count_repetition_guard": "9", + "avg_total_tokens_target_model": "11154.00", + "total_total_tokens_target_model": "100386.0", + "avg_total_tokens_other_models": "57830.22", + "total_total_tokens_other_models": "520472.0", + "avg_total_tokens_all_models": "68984.22", + "total_total_tokens_all_models": "620858.0", + "avg_turn_count": "26.78", + "total_turn_count": "241.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/36a174ea for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "9", + "sample_ids": "20cdebb7-9189-4f09-ad21-bfc8e8b75680,6f827e79-05c4-4383-8ed4-16d36260f29f,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776784808.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.45674974470724616, + "uncertainty": { + "num_samples": 36 + }, + "details": { + "total_matched_trajectories": "9", + "stopping_reason_count_repetition_guard": "9", + "avg_total_tokens_target_model": "11154.00", + "total_total_tokens_target_model": "100386.0", + "avg_total_tokens_other_models": "57830.22", + "total_total_tokens_other_models": "520472.0", + "avg_total_tokens_all_models": "68984.22", + "total_total_tokens_all_models": "620858.0", + "avg_turn_count": "26.78", + "total_turn_count": "241.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/89899de3-59ad-41c3-8b09-63f587910abe_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "8ae90f213fa454e78ccc90273c07ab71893f5d6bbc248157d084383cbb953c07", + "total_rows": 36 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/89899de3-59ad-41c3-8b09-63f587910abe_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/89899de3-59ad-41c3-8b09-63f587910abe_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..7358036045e9581bcc3d593aaab26ce53e73f0c3 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/89899de3-59ad-41c3-8b09-63f587910abe_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ae90f213fa454e78ccc90273c07ab71893f5d6bbc248157d084383cbb953c07 +size 2431630 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/8ae4e2f7-8043-4b4b-b5e3-a175939b8d8d.json b/data/healthbench/openai/gpt-5.2-2025-12-11/8ae4e2f7-8043-4b4b-b5e3-a175939b8d8d.json new file mode 100644 index 0000000000000000000000000000000000000000..3891e06d8814000b57e031e2add1a32077b02b7a --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/8ae4e2f7-8043-4b4b-b5e3-a175939b8d8d.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117841.030177", + "evaluation_timestamp": "1776165242.0", + "retrieved_timestamp": "1786117841.030177", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+3ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "34", + "sample_ids": "9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4443969998832628, + "uncertainty": { + "standard_deviation": 0.03061040288246589, + "num_samples": 102 + }, + "details": { + "total_matched_trajectories": "15", + "stopping_reason_count_repetition_guard": "15", + "avg_total_tokens_target_model": "10572.07", + "total_total_tokens_target_model": "158581.0", + "avg_total_tokens_other_models": "55800.93", + "total_total_tokens_other_models": "837014.0", + "avg_total_tokens_all_models": "66373.00", + "total_total_tokens_all_models": "995595.0", + "avg_turn_count": "26.73", + "total_turn_count": "401.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+3ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "34", + "sample_ids": "9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.03061040288246589, + "uncertainty": { + "standard_deviation": 0.03061040288246589, + "num_samples": 102 + }, + "details": { + "total_matched_trajectories": "15", + "stopping_reason_count_repetition_guard": "15", + "avg_total_tokens_target_model": "10572.07", + "total_total_tokens_target_model": "158581.0", + "avg_total_tokens_other_models": "55800.93", + "total_total_tokens_other_models": "837014.0", + "avg_total_tokens_all_models": "66373.00", + "total_total_tokens_all_models": "995595.0", + "avg_turn_count": "26.73", + "total_turn_count": "401.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+3ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "34", + "sample_ids": "9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.46128769356153215, + "uncertainty": { + "standard_deviation": 0.03061040288246589, + "num_samples": 102 + }, + "details": { + "total_matched_trajectories": "15", + "stopping_reason_count_repetition_guard": "15", + "avg_total_tokens_target_model": "10572.07", + "total_total_tokens_target_model": "158581.0", + "avg_total_tokens_other_models": "55800.93", + "total_total_tokens_other_models": "837014.0", + "avg_total_tokens_all_models": "66373.00", + "total_total_tokens_all_models": "995595.0", + "avg_turn_count": "26.73", + "total_turn_count": "401.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+3ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "34", + "sample_ids": "9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9" + } + }, + "evaluation_timestamp": "1776165242.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4443969998832628, + "uncertainty": { + "num_samples": 102 + }, + "details": { + "total_matched_trajectories": "15", + "stopping_reason_count_repetition_guard": "15", + "avg_total_tokens_target_model": "10572.07", + "total_total_tokens_target_model": "158581.0", + "avg_total_tokens_other_models": "55800.93", + "total_total_tokens_other_models": "837014.0", + "avg_total_tokens_all_models": "66373.00", + "total_total_tokens_all_models": "995595.0", + "avg_turn_count": "26.73", + "total_turn_count": "401.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/8ae4e2f7-8043-4b4b-b5e3-a175939b8d8d_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "02226ecd1e1025c544c2dcda4241bdda2fb725642fed5dae606be54f857d6327", + "total_rows": 102 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/8ae4e2f7-8043-4b4b-b5e3-a175939b8d8d_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/8ae4e2f7-8043-4b4b-b5e3-a175939b8d8d_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..f5d3695f51d1d57b5f29d6da5e747d1e9143d74e --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/8ae4e2f7-8043-4b4b-b5e3-a175939b8d8d_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02226ecd1e1025c544c2dcda4241bdda2fb725642fed5dae606be54f857d6327 +size 6554750 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/a0a24c9c-a71c-4b2f-aed7-589cf81512e5.json b/data/healthbench/openai/gpt-5.2-2025-12-11/a0a24c9c-a71c-4b2f-aed7-589cf81512e5.json new file mode 100644 index 0000000000000000000000000000000000000000..127881def8ac8369a9a77319fe8f582dc83a05de --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/a0a24c9c-a71c-4b2f-aed7-589cf81512e5.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117864.040277", + "evaluation_timestamp": "1776259833.0", + "retrieved_timestamp": "1786117864.040277", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+3ep/47b9ab59 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.516656183656474, + "uncertainty": { + "standard_deviation": 0.0467594358818376, + "num_samples": 111 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "12802.67", + "total_total_tokens_target_model": "153632.0", + "avg_total_tokens_other_models": "74340.67", + "total_total_tokens_other_models": "892088.0", + "avg_total_tokens_all_models": "87143.33", + "total_total_tokens_all_models": "1045720.0", + "avg_turn_count": "32.75", + "total_turn_count": "393.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+3ep/47b9ab59 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0467594358818376, + "uncertainty": { + "standard_deviation": 0.0467594358818376, + "num_samples": 111 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "12802.67", + "total_total_tokens_target_model": "153632.0", + "avg_total_tokens_other_models": "74340.67", + "total_total_tokens_other_models": "892088.0", + "avg_total_tokens_all_models": "87143.33", + "total_total_tokens_all_models": "1045720.0", + "avg_turn_count": "32.75", + "total_turn_count": "393.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+3ep/47b9ab59 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.49210307564422273, + "uncertainty": { + "standard_deviation": 0.0467594358818376, + "num_samples": 111 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "12802.67", + "total_total_tokens_target_model": "153632.0", + "avg_total_tokens_other_models": "74340.67", + "total_total_tokens_other_models": "892088.0", + "avg_total_tokens_all_models": "87143.33", + "total_total_tokens_all_models": "1045720.0", + "avg_turn_count": "32.75", + "total_turn_count": "393.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+3ep/47b9ab59 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d,c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.516656183656474, + "uncertainty": { + "num_samples": 111 + }, + "details": { + "total_matched_trajectories": "12", + "stopping_reason_count_repetition_guard": "12", + "avg_total_tokens_target_model": "12802.67", + "total_total_tokens_target_model": "153632.0", + "avg_total_tokens_other_models": "74340.67", + "total_total_tokens_other_models": "892088.0", + "avg_total_tokens_all_models": "87143.33", + "total_total_tokens_all_models": "1045720.0", + "avg_turn_count": "32.75", + "total_turn_count": "393.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/a0a24c9c-a71c-4b2f-aed7-589cf81512e5_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "acb02df9a3ff363c3a0eca2a2c362c257062a7f93fd863bd829250eb024b09e3", + "total_rows": 111 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/a0a24c9c-a71c-4b2f-aed7-589cf81512e5_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/a0a24c9c-a71c-4b2f-aed7-589cf81512e5_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..09bcba4597cb0edc8f70b80c7afcbfd6b9b51cf1 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/a0a24c9c-a71c-4b2f-aed7-589cf81512e5_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:acb02df9a3ff363c3a0eca2a2c362c257062a7f93fd863bd829250eb024b09e3 +size 6706108 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/ba3f4ed7-2789-4d85-9f4e-d08dc54ddbd6.json b/data/healthbench/openai/gpt-5.2-2025-12-11/ba3f4ed7-2789-4d85-9f4e-d08dc54ddbd6.json new file mode 100644 index 0000000000000000000000000000000000000000..ea9a203bf90f32534ce03867360034b1c503045d --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/ba3f4ed7-2789-4d85-9f4e-d08dc54ddbd6.json @@ -0,0 +1,322 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117820.358133", + "evaluation_timestamp": "1776779722.0", + "retrieved_timestamp": "1786117820.358133", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep/dd4052a1 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "27", + "sample_ids": "269074dc-9495-484b-b2f9-0f0943c0f816,1dc0b953-ef66-4a78-9989-7cf1053b5b41,3cf35bfe-d881-4881-a268-b5fa0607bf89,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,99b22b39-30d5-40dc-82d9-c1cd3c793a44,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.31550374605702414, + "uncertainty": { + "standard_deviation": 0.04633381376400383, + "num_samples": 53 + }, + "details": { + "total_matched_trajectories": "53", + "stopping_reason_count_repetition_guard": "52", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "11966.96", + "total_total_tokens_target_model": "634249.0", + "avg_total_tokens_other_models": "59470.87", + "total_total_tokens_other_models": "3151956.0", + "avg_total_tokens_all_models": "71437.83", + "total_total_tokens_all_models": "3786205.0", + "avg_turn_count": "28.15", + "total_turn_count": "1492.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep/dd4052a1 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "27", + "sample_ids": "269074dc-9495-484b-b2f9-0f0943c0f816,1dc0b953-ef66-4a78-9989-7cf1053b5b41,3cf35bfe-d881-4881-a268-b5fa0607bf89,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,99b22b39-30d5-40dc-82d9-c1cd3c793a44,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.04633381376400383, + "uncertainty": { + "standard_deviation": 0.04633381376400383, + "num_samples": 53 + }, + "details": { + "total_matched_trajectories": "53", + "stopping_reason_count_repetition_guard": "52", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "11966.96", + "total_total_tokens_target_model": "634249.0", + "avg_total_tokens_other_models": "59470.87", + "total_total_tokens_other_models": "3151956.0", + "avg_total_tokens_all_models": "71437.83", + "total_total_tokens_all_models": "3786205.0", + "avg_turn_count": "28.15", + "total_turn_count": "1492.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep/dd4052a1 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "27", + "sample_ids": "269074dc-9495-484b-b2f9-0f0943c0f816,1dc0b953-ef66-4a78-9989-7cf1053b5b41,3cf35bfe-d881-4881-a268-b5fa0607bf89,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,99b22b39-30d5-40dc-82d9-c1cd3c793a44,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.39906832298136646, + "uncertainty": { + "standard_deviation": 0.04633381376400383, + "num_samples": 53 + }, + "details": { + "total_matched_trajectories": "53", + "stopping_reason_count_repetition_guard": "52", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "11966.96", + "total_total_tokens_target_model": "634249.0", + "avg_total_tokens_other_models": "59470.87", + "total_total_tokens_other_models": "3151956.0", + "avg_total_tokens_all_models": "71437.83", + "total_total_tokens_all_models": "3786205.0", + "avg_turn_count": "28.15", + "total_turn_count": "1492.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep/dd4052a1 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "27", + "sample_ids": "269074dc-9495-484b-b2f9-0f0943c0f816,1dc0b953-ef66-4a78-9989-7cf1053b5b41,3cf35bfe-d881-4881-a268-b5fa0607bf89,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,99b22b39-30d5-40dc-82d9-c1cd3c793a44,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,48c89cd1-0ecf-4738-8fcc-392269585bd3,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,36c6e780-d4a8-47b2-8a36-90ee4b97607f,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776779722.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.31550374605702414, + "uncertainty": { + "num_samples": 53 + }, + "details": { + "total_matched_trajectories": "53", + "stopping_reason_count_repetition_guard": "52", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "11966.96", + "total_total_tokens_target_model": "634249.0", + "avg_total_tokens_other_models": "59470.87", + "total_total_tokens_other_models": "3151956.0", + "avg_total_tokens_all_models": "71437.83", + "total_total_tokens_all_models": "3786205.0", + "avg_turn_count": "28.15", + "total_turn_count": "1492.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/ba3f4ed7-2789-4d85-9f4e-d08dc54ddbd6_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "a427cbeefdaf2d9d7f5cacb19d1250cd869722247f66459d805bbbcfda6242eb", + "total_rows": 53 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/ba3f4ed7-2789-4d85-9f4e-d08dc54ddbd6_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/ba3f4ed7-2789-4d85-9f4e-d08dc54ddbd6_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..f8b0c5ffc9adaac874a5ed419c1a703e41817776 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/ba3f4ed7-2789-4d85-9f4e-d08dc54ddbd6_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a427cbeefdaf2d9d7f5cacb19d1250cd869722247f66459d805bbbcfda6242eb +size 3925262 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/bf47c82c-151f-4f7e-8103-c92d777fb709.json b/data/healthbench/openai/gpt-5.2-2025-12-11/bf47c82c-151f-4f7e-8103-c92d777fb709.json new file mode 100644 index 0000000000000000000000000000000000000000..f308577dca6668f57022b8b66e462aa25ec3930d --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/bf47c82c-151f-4f7e-8103-c92d777fb709.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117867.608804", + "evaluation_timestamp": "1776278452.0", + "retrieved_timestamp": "1786117867.608804", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/25510848 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2" + } + }, + "evaluation_timestamp": "1776278452.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.42441417814109317, + "uncertainty": { + "standard_deviation": 0.1169134724738184, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "9941.38", + "total_total_tokens_target_model": "79531.0", + "avg_total_tokens_other_models": "46314.50", + "total_total_tokens_other_models": "370516.0", + "avg_total_tokens_all_models": "56255.88", + "total_total_tokens_all_models": "450047.0", + "avg_turn_count": "21.12", + "total_turn_count": "169.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/25510848 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2" + } + }, + "evaluation_timestamp": "1776278452.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1169134724738184, + "uncertainty": { + "standard_deviation": 0.1169134724738184, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "9941.38", + "total_total_tokens_target_model": "79531.0", + "avg_total_tokens_other_models": "46314.50", + "total_total_tokens_other_models": "370516.0", + "avg_total_tokens_all_models": "56255.88", + "total_total_tokens_all_models": "450047.0", + "avg_turn_count": "21.12", + "total_turn_count": "169.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/25510848 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2" + } + }, + "evaluation_timestamp": "1776278452.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.578125, + "uncertainty": { + "standard_deviation": 0.1169134724738184, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "9941.38", + "total_total_tokens_target_model": "79531.0", + "avg_total_tokens_other_models": "46314.50", + "total_total_tokens_other_models": "370516.0", + "avg_total_tokens_all_models": "56255.88", + "total_total_tokens_all_models": "450047.0", + "avg_turn_count": "21.12", + "total_turn_count": "169.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/25510848 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2" + } + }, + "evaluation_timestamp": "1776278452.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.42441417814109317, + "uncertainty": { + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "8", + "avg_total_tokens_target_model": "9941.38", + "total_total_tokens_target_model": "79531.0", + "avg_total_tokens_other_models": "46314.50", + "total_total_tokens_other_models": "370516.0", + "avg_total_tokens_all_models": "56255.88", + "total_total_tokens_all_models": "450047.0", + "avg_turn_count": "21.12", + "total_turn_count": "169.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/bf47c82c-151f-4f7e-8103-c92d777fb709_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "31a34a8ef7e8841094edd8b868c4e3d6d9d95bd68d930328fed1737c74934d9d", + "total_rows": 40 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/bf47c82c-151f-4f7e-8103-c92d777fb709_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/bf47c82c-151f-4f7e-8103-c92d777fb709_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c86623e292d062880740a25e74b9d2cfbc919da7 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/bf47c82c-151f-4f7e-8103-c92d777fb709_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31a34a8ef7e8841094edd8b868c4e3d6d9d95bd68d930328fed1737c74934d9d +size 2670270 diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/e6850e09-4a37-4c6b-891c-c73144806aca.json b/data/healthbench/openai/gpt-5.2-2025-12-11/e6850e09-4a37-4c6b-891c-c73144806aca.json new file mode 100644 index 0000000000000000000000000000000000000000..2f63475a26b4ec757dca4a18b05dcea007c67ac5 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/e6850e09-4a37-4c6b-891c-c73144806aca.json @@ -0,0 +1,322 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.2-2025-12-11/1786117874.477024", + "evaluation_timestamp": "1776711165.0", + "retrieved_timestamp": "1786117874.477024", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/ce64842b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "20cdebb7-9189-4f09-ad21-bfc8e8b75680,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776711165.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3902289746266607, + "uncertainty": { + "standard_deviation": 0.0667239959327993, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "40", + "stopping_reason_count_repetition_guard": "38", + "stopping_reason_count_completed_with_submit": "2", + "avg_total_tokens_target_model": "11087.95", + "total_total_tokens_target_model": "443518.0", + "avg_total_tokens_other_models": "60557.55", + "total_total_tokens_other_models": "2422302.0", + "avg_total_tokens_all_models": "71645.50", + "total_total_tokens_all_models": "2865820.0", + "avg_turn_count": "27.18", + "total_turn_count": "1087.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/ce64842b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "20cdebb7-9189-4f09-ad21-bfc8e8b75680,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776711165.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0667239959327993, + "uncertainty": { + "standard_deviation": 0.0667239959327993, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "40", + "stopping_reason_count_repetition_guard": "38", + "stopping_reason_count_completed_with_submit": "2", + "avg_total_tokens_target_model": "11087.95", + "total_total_tokens_target_model": "443518.0", + "avg_total_tokens_other_models": "60557.55", + "total_total_tokens_other_models": "2422302.0", + "avg_total_tokens_all_models": "71645.50", + "total_total_tokens_all_models": "2865820.0", + "avg_turn_count": "27.18", + "total_turn_count": "1087.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/ce64842b for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "20cdebb7-9189-4f09-ad21-bfc8e8b75680,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776711165.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4180327868852459, + "uncertainty": { + "standard_deviation": 0.0667239959327993, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "40", + "stopping_reason_count_repetition_guard": "38", + "stopping_reason_count_completed_with_submit": "2", + "avg_total_tokens_target_model": "11087.95", + "total_total_tokens_target_model": "443518.0", + "avg_total_tokens_other_models": "60557.55", + "total_total_tokens_other_models": "2422302.0", + "avg_total_tokens_all_models": "71645.50", + "total_total_tokens_all_models": "2865820.0", + "avg_turn_count": "27.18", + "total_turn_count": "1087.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/ce64842b for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "10", + "sample_ids": "20cdebb7-9189-4f09-ad21-bfc8e8b75680,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776711165.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3902289746266607, + "uncertainty": { + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "40", + "stopping_reason_count_repetition_guard": "38", + "stopping_reason_count_completed_with_submit": "2", + "avg_total_tokens_target_model": "11087.95", + "total_total_tokens_target_model": "443518.0", + "avg_total_tokens_other_models": "60557.55", + "total_total_tokens_other_models": "2422302.0", + "avg_total_tokens_all_models": "71645.50", + "total_total_tokens_all_models": "2865820.0", + "avg_turn_count": "27.18", + "total_turn_count": "1087.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.2-2025-12-11/e6850e09-4a37-4c6b-891c-c73144806aca_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "1f157846a5af6619bb4cc0f75a560599fbc1bf4ac5b04d29b47f6533ba6b5560", + "total_rows": 40 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.2-2025-12-11/e6850e09-4a37-4c6b-891c-c73144806aca_samples.jsonl b/data/healthbench/openai/gpt-5.2-2025-12-11/e6850e09-4a37-4c6b-891c-c73144806aca_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..fd6fac18202aa11c4cc22fc89e991d915ed6acb0 --- /dev/null +++ b/data/healthbench/openai/gpt-5.2-2025-12-11/e6850e09-4a37-4c6b-891c-c73144806aca_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f157846a5af6619bb4cc0f75a560599fbc1bf4ac5b04d29b47f6533ba6b5560 +size 2954066 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/0f057841-1025-4d08-89ee-00c45ea64c58.json b/data/healthbench/openai/gpt-5.4-2026-03-05/0f057841-1025-4d08-89ee-00c45ea64c58.json new file mode 100644 index 0000000000000000000000000000000000000000..974292593af00c41b2078e52fba97fc4909747be --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/0f057841-1025-4d08-89ee-00c45ea64c58.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786117997.401406", + "evaluation_timestamp": "1775558527.0", + "retrieved_timestamp": "1786117997.401406", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6133256647098454, + "uncertainty": { + "standard_deviation": 0.1106389674410679, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "61480.00", + "total_total_tokens_target_model": "61480.0", + "avg_total_tokens_other_models": "117747.00", + "total_total_tokens_other_models": "117747.0", + "avg_total_tokens_all_models": "179227.00", + "total_total_tokens_all_models": "179227.0", + "avg_turn_count": "23.00", + "total_turn_count": "23.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1106389674410679, + "uncertainty": { + "standard_deviation": 0.1106389674410679, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "61480.00", + "total_total_tokens_target_model": "61480.0", + "avg_total_tokens_other_models": "117747.00", + "total_total_tokens_other_models": "117747.0", + "avg_total_tokens_all_models": "179227.00", + "total_total_tokens_all_models": "179227.0", + "avg_turn_count": "23.00", + "total_turn_count": "23.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.552734375, + "uncertainty": { + "standard_deviation": 0.1106389674410679, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "61480.00", + "total_total_tokens_target_model": "61480.0", + "avg_total_tokens_other_models": "117747.00", + "total_total_tokens_other_models": "117747.0", + "avg_total_tokens_all_models": "179227.00", + "total_total_tokens_all_models": "179227.0", + "avg_turn_count": "23.00", + "total_turn_count": "23.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+8ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6133256647098454, + "uncertainty": { + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "61480.00", + "total_total_tokens_target_model": "61480.0", + "avg_total_tokens_other_models": "117747.00", + "total_total_tokens_other_models": "117747.0", + "avg_total_tokens_all_models": "179227.00", + "total_total_tokens_all_models": "179227.0", + "avg_turn_count": "23.00", + "total_turn_count": "23.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/0f057841-1025-4d08-89ee-00c45ea64c58_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "5d8829980a8adac24559dc2f124f4d6c3ab221dafec9df0df99cc230dc060cfb", + "total_rows": 40 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/0f057841-1025-4d08-89ee-00c45ea64c58_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/0f057841-1025-4d08-89ee-00c45ea64c58_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b8d90f2e13db3b29fc0fabdf92375f2b44fc6f62 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/0f057841-1025-4d08-89ee-00c45ea64c58_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d8829980a8adac24559dc2f124f4d6c3ab221dafec9df0df99cc230dc060cfb +size 2930414 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/1d34792d-9d34-4a12-899b-439808383e9e.json b/data/healthbench/openai/gpt-5.4-2026-03-05/1d34792d-9d34-4a12-899b-439808383e9e.json new file mode 100644 index 0000000000000000000000000000000000000000..1da9921dcd11f499218ee82080a272983842ea67 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/1d34792d-9d34-4a12-899b-439808383e9e.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786117938.804666", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786117938.804666", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6189591137048764, + "uncertainty": { + "standard_deviation": 0.11205785678908674, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "49654.00", + "total_total_tokens_target_model": "99308.0", + "avg_total_tokens_other_models": "101726.50", + "total_total_tokens_other_models": "203453.0", + "avg_total_tokens_all_models": "151380.50", + "total_total_tokens_all_models": "302761.0", + "avg_turn_count": "23.00", + "total_turn_count": "46.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.11205785678908674, + "uncertainty": { + "standard_deviation": 0.11205785678908674, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "49654.00", + "total_total_tokens_target_model": "99308.0", + "avg_total_tokens_other_models": "101726.50", + "total_total_tokens_other_models": "203453.0", + "avg_total_tokens_all_models": "151380.50", + "total_total_tokens_all_models": "302761.0", + "avg_turn_count": "23.00", + "total_turn_count": "46.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.578125, + "uncertainty": { + "standard_deviation": 0.11205785678908674, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "49654.00", + "total_total_tokens_target_model": "99308.0", + "avg_total_tokens_other_models": "101726.50", + "total_total_tokens_other_models": "203453.0", + "avg_total_tokens_all_models": "151380.50", + "total_total_tokens_all_models": "302761.0", + "avg_turn_count": "23.00", + "total_turn_count": "46.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6189591137048764, + "uncertainty": { + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "49654.00", + "total_total_tokens_target_model": "99308.0", + "avg_total_tokens_other_models": "101726.50", + "total_total_tokens_other_models": "203453.0", + "avg_total_tokens_all_models": "151380.50", + "total_total_tokens_all_models": "302761.0", + "avg_turn_count": "23.00", + "total_turn_count": "46.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/1d34792d-9d34-4a12-899b-439808383e9e_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "7985697e5121b16c9273564731f2e614445f5465b808243f9f6bf256d239c2d7", + "total_rows": 10 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/1d34792d-9d34-4a12-899b-439808383e9e_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/1d34792d-9d34-4a12-899b-439808383e9e_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..e7b08820b566e0ec9e90b35fb01e678e2605b299 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/1d34792d-9d34-4a12-899b-439808383e9e_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7985697e5121b16c9273564731f2e614445f5465b808243f9f6bf256d239c2d7 +size 745024 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/32c70c6a-e55b-499b-964d-1765489ab933.json b/data/healthbench/openai/gpt-5.4-2026-03-05/32c70c6a-e55b-499b-964d-1765489ab933.json new file mode 100644 index 0000000000000000000000000000000000000000..da016473d0eede439a208473b708b982664cdd8c --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/32c70c6a-e55b-499b-964d-1765489ab933.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786111990.170989", + "evaluation_timestamp": "1775558527.0", + "retrieved_timestamp": "1786111990.170989", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6133256647098454, + "uncertainty": { + "standard_deviation": 0.1106389674410679, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "61480.00", + "total_total_tokens_target_model": "61480.0", + "avg_total_tokens_other_models": "117747.00", + "total_total_tokens_other_models": "117747.0", + "avg_total_tokens_all_models": "179227.00", + "total_total_tokens_all_models": "179227.0", + "avg_turn_count": "23.00", + "total_turn_count": "23.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1106389674410679, + "uncertainty": { + "standard_deviation": 0.1106389674410679, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "61480.00", + "total_total_tokens_target_model": "61480.0", + "avg_total_tokens_other_models": "117747.00", + "total_total_tokens_other_models": "117747.0", + "avg_total_tokens_all_models": "179227.00", + "total_total_tokens_all_models": "179227.0", + "avg_turn_count": "23.00", + "total_turn_count": "23.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+8ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.552734375, + "uncertainty": { + "standard_deviation": 0.1106389674410679, + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "61480.00", + "total_total_tokens_target_model": "61480.0", + "avg_total_tokens_other_models": "117747.00", + "total_total_tokens_other_models": "117747.0", + "avg_total_tokens_all_models": "179227.00", + "total_total_tokens_all_models": "179227.0", + "avg_turn_count": "23.00", + "total_turn_count": "23.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+8ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775558527.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6133256647098454, + "uncertainty": { + "num_samples": 40 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "61480.00", + "total_total_tokens_target_model": "61480.0", + "avg_total_tokens_other_models": "117747.00", + "total_total_tokens_other_models": "117747.0", + "avg_total_tokens_all_models": "179227.00", + "total_total_tokens_all_models": "179227.0", + "avg_turn_count": "23.00", + "total_turn_count": "23.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/32c70c6a-e55b-499b-964d-1765489ab933_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "5d8829980a8adac24559dc2f124f4d6c3ab221dafec9df0df99cc230dc060cfb", + "total_rows": 40 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/32c70c6a-e55b-499b-964d-1765489ab933_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/32c70c6a-e55b-499b-964d-1765489ab933_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b8d90f2e13db3b29fc0fabdf92375f2b44fc6f62 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/32c70c6a-e55b-499b-964d-1765489ab933_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d8829980a8adac24559dc2f124f4d6c3ab221dafec9df0df99cc230dc060cfb +size 2930414 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/4ccac2ec-4abf-4e3f-8824-6b9473a4f29d.json b/data/healthbench/openai/gpt-5.4-2026-03-05/4ccac2ec-4abf-4e3f-8824-6b9473a4f29d.json new file mode 100644 index 0000000000000000000000000000000000000000..00d9785153f2b4cc32f114e998c170881083bec7 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/4ccac2ec-4abf-4e3f-8824-6b9473a4f29d.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786117912.401338", + "evaluation_timestamp": "1776781177.0", + "retrieved_timestamp": "1786117912.401338", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+1ep/c7e069f5 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "be6150a5-8060-4434-9724-afede307d495,a6204079-9e72-42ad-8f39-a65552fa2e68,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,99b22b39-30d5-40dc-82d9-c1cd3c793a44,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5" + } + }, + "evaluation_timestamp": "1776781177.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.23085825026183682, + "uncertainty": { + "standard_deviation": 0.043076215499593014, + "num_samples": 49 + }, + "details": { + "total_matched_trajectories": "49", + "stopping_reason_count_repetition_guard": "49", + "avg_total_tokens_target_model": "15151.35", + "total_total_tokens_target_model": "742416.0", + "avg_total_tokens_other_models": "61073.08", + "total_total_tokens_other_models": "2992581.0", + "avg_total_tokens_all_models": "76224.43", + "total_total_tokens_all_models": "3734997.0", + "avg_turn_count": "26.31", + "total_turn_count": "1289.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+1ep/c7e069f5 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "be6150a5-8060-4434-9724-afede307d495,a6204079-9e72-42ad-8f39-a65552fa2e68,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,99b22b39-30d5-40dc-82d9-c1cd3c793a44,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5" + } + }, + "evaluation_timestamp": "1776781177.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.043076215499593014, + "uncertainty": { + "standard_deviation": 0.043076215499593014, + "num_samples": 49 + }, + "details": { + "total_matched_trajectories": "49", + "stopping_reason_count_repetition_guard": "49", + "avg_total_tokens_target_model": "15151.35", + "total_total_tokens_target_model": "742416.0", + "avg_total_tokens_other_models": "61073.08", + "total_total_tokens_other_models": "2992581.0", + "avg_total_tokens_all_models": "76224.43", + "total_total_tokens_all_models": "3734997.0", + "avg_turn_count": "26.31", + "total_turn_count": "1289.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+1ep/c7e069f5 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "be6150a5-8060-4434-9724-afede307d495,a6204079-9e72-42ad-8f39-a65552fa2e68,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,99b22b39-30d5-40dc-82d9-c1cd3c793a44,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5" + } + }, + "evaluation_timestamp": "1776781177.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3087719298245614, + "uncertainty": { + "standard_deviation": 0.043076215499593014, + "num_samples": 49 + }, + "details": { + "total_matched_trajectories": "49", + "stopping_reason_count_repetition_guard": "49", + "avg_total_tokens_target_model": "15151.35", + "total_total_tokens_target_model": "742416.0", + "avg_total_tokens_other_models": "61073.08", + "total_total_tokens_other_models": "2992581.0", + "avg_total_tokens_all_models": "76224.43", + "total_total_tokens_all_models": "3734997.0", + "avg_turn_count": "26.31", + "total_turn_count": "1289.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+1ep/c7e069f5 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "50", + "sample_ids": "be6150a5-8060-4434-9724-afede307d495,a6204079-9e72-42ad-8f39-a65552fa2e68,3dc41419-2565-428e-b9b9-02e3fd93c470,a91a22f8-d2e6-4593-bcae-6d479ed38031,d0d3426f-8b7a-4501-adb4-86d8fa9099c8,a7dac6a9-eab5-4c01-882c-dbf6c774d7c3,c1240427-ae38-4335-9e80-492c13ef4893,b7ddde65-702e-4255-b1d7-09d01248e76d,00d34549-134c-48e2-b4bb-b54946da70cc,85c11e2a-628b-464f-a8bb-e97ab04a73eb,69a544f8-957d-4dda-a777-608ec7b95617,90789a49-ab1f-4229-bd89-e1c56ea8be3a,cd53e21a-e5ed-4d9b-a258-42a23e9715bb,8d409c7b-29d2-4df2-aa69-ab56c9339bc5,6566f489-f254-4bca-a86b-90f5ac7471d7,d9a44a36-6a3c-4d5d-9c66-b2c0afb479d1,9cf87956-37e6-4bed-9cea-a64b13ae0de5,1dc0b953-ef66-4a78-9989-7cf1053b5b41,a1a08e1d-1353-4d6f-94ce-f90872074c7c,51d9c72b-ba60-47c5-9cc9-3ed6212a3982,3cf35bfe-d881-4881-a268-b5fa0607bf89,bc8d3e4a-8ba0-4cab-be1c-b20ae3af91e4,99306c29-bcef-49d5-bcd1-78d4e541e197,d0fcfaf0-7c6d-4ead-915b-6d9d9130e747,59c68d10-55c1-4629-b05a-bedde7f343b5,d86ee090-3a2e-44d0-abed-6f0f40a38c08,25cf07b8-606f-4448-99ef-ea87ca3097ba,1ca222ed-cc79-4e78-9ca3-3547e7b37e3a,63e4e0d4-0cb1-442e-8316-8c12a6e3ec14,0eb6fd6d-eaaa-46db-ab16-02e610f238a9,99b22b39-30d5-40dc-82d9-c1cd3c793a44,8c342edc-619b-4079-a28c-b7129f3670a6,e8991792-c875-4f40-99a0-54c902d42a52,54349c88-e80a-4728-b181-305f97905d59,f93b0850-2c60-479d-831b-b8c8f9493d61,c63a4306-4003-45a6-8be1-d669457473a9,d3cbb579-70a5-46e8-95f6-72cc9afa4ba3,3852d050-c009-4e33-b953-fd734a5b5e73,5dc23430-dfbb-4794-9f77-0d4dd6093033,67e76447-f16a-4a65-aba7-fc7166fd9b63,a783e7a3-cecc-4d15-b18c-3f2ddfc2dc6c,62e9d35f-153d-4744-b97c-9abfb2a90b1e,04b6e0c8-2835-415e-865d-aca9f1cd3c57,7a283823-00ee-489d-8dbc-356e2857526f,e89e00c9-29e9-420c-813d-e06ddc39d585,51433193-b0c3-4725-a3a8-9dd15e969b0c,92ea478e-6bf8-4ffb-b88a-f57ccf5f2346,5fb6812f-9816-4d5d-b7e0-4ec3e9df5dc1,c29c95d1-5b5e-4f53-9ec5-83e1f72b5aa5,b81c059a-1137-47cf-bab8-ab5eb8ff34d5" + } + }, + "evaluation_timestamp": "1776781177.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.23085825026183682, + "uncertainty": { + "num_samples": 49 + }, + "details": { + "total_matched_trajectories": "49", + "stopping_reason_count_repetition_guard": "49", + "avg_total_tokens_target_model": "15151.35", + "total_total_tokens_target_model": "742416.0", + "avg_total_tokens_other_models": "61073.08", + "total_total_tokens_other_models": "2992581.0", + "avg_total_tokens_all_models": "76224.43", + "total_total_tokens_all_models": "3734997.0", + "avg_turn_count": "26.31", + "total_turn_count": "1289.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/4ccac2ec-4abf-4e3f-8824-6b9473a4f29d_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "f4119a96f46fcaf3bac3eb86dd00fa981961fe1ccd6b1a0892a2ce07cc86e06a", + "total_rows": 49 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/4ccac2ec-4abf-4e3f-8824-6b9473a4f29d_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/4ccac2ec-4abf-4e3f-8824-6b9473a4f29d_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..8cad2d821cb898d8579c41b3d765da3f416e36b0 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/4ccac2ec-4abf-4e3f-8824-6b9473a4f29d_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4119a96f46fcaf3bac3eb86dd00fa981961fe1ccd6b1a0892a2ce07cc86e06a +size 3466512 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/6b177a5b-58a7-4ac2-b277-b67df1d5604f.json b/data/healthbench/openai/gpt-5.4-2026-03-05/6b177a5b-58a7-4ac2-b277-b67df1d5604f.json new file mode 100644 index 0000000000000000000000000000000000000000..a146801f0e183b763c8cdbacaef801b11214686e --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/6b177a5b-58a7-4ac2-b277-b67df1d5604f.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786117954.130157", + "evaluation_timestamp": "1776259833.0", + "retrieved_timestamp": "1786117954.130157", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep/9913b0c4 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "60", + "sample_ids": "77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e,3bcc3abf-5e76-4d3d-b0bb-c74cd0e5f4d8,c3eeecfc-de96-480f-929d-ac4a5e5a541c,3b9140da-0643-40b9-a610-fe7657942509,e6866a00-ab62-4e5d-9953-f0178ea3e570,9b731b02-db03-484d-a06e-1b55a4df57b8" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4236937487216529, + "uncertainty": { + "standard_deviation": 0.036847150788597155, + "num_samples": 120 + }, + "details": { + "total_matched_trajectories": "24", + "stopping_reason_count_repetition_guard": "24", + "avg_total_tokens_target_model": "14460.25", + "total_total_tokens_target_model": "347046.0", + "avg_total_tokens_other_models": "59433.04", + "total_total_tokens_other_models": "1426393.0", + "avg_total_tokens_all_models": "73893.29", + "total_total_tokens_all_models": "1773439.0", + "avg_turn_count": "26.21", + "total_turn_count": "629.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep/9913b0c4 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "60", + "sample_ids": "77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e,3bcc3abf-5e76-4d3d-b0bb-c74cd0e5f4d8,c3eeecfc-de96-480f-929d-ac4a5e5a541c,3b9140da-0643-40b9-a610-fe7657942509,e6866a00-ab62-4e5d-9953-f0178ea3e570,9b731b02-db03-484d-a06e-1b55a4df57b8" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.036847150788597155, + "uncertainty": { + "standard_deviation": 0.036847150788597155, + "num_samples": 120 + }, + "details": { + "total_matched_trajectories": "24", + "stopping_reason_count_repetition_guard": "24", + "avg_total_tokens_target_model": "14460.25", + "total_total_tokens_target_model": "347046.0", + "avg_total_tokens_other_models": "59433.04", + "total_total_tokens_other_models": "1426393.0", + "avg_total_tokens_all_models": "73893.29", + "total_total_tokens_all_models": "1773439.0", + "avg_turn_count": "26.21", + "total_turn_count": "629.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep/9913b0c4 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "60", + "sample_ids": "77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e,3bcc3abf-5e76-4d3d-b0bb-c74cd0e5f4d8,c3eeecfc-de96-480f-929d-ac4a5e5a541c,3b9140da-0643-40b9-a610-fe7657942509,e6866a00-ab62-4e5d-9953-f0178ea3e570,9b731b02-db03-484d-a06e-1b55a4df57b8" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4270833333333333, + "uncertainty": { + "standard_deviation": 0.036847150788597155, + "num_samples": 120 + }, + "details": { + "total_matched_trajectories": "24", + "stopping_reason_count_repetition_guard": "24", + "avg_total_tokens_target_model": "14460.25", + "total_total_tokens_target_model": "347046.0", + "avg_total_tokens_other_models": "59433.04", + "total_total_tokens_other_models": "1426393.0", + "avg_total_tokens_all_models": "73893.29", + "total_total_tokens_all_models": "1773439.0", + "avg_turn_count": "26.21", + "total_turn_count": "629.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep/9913b0c4 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "60", + "sample_ids": "77c0be2e-364a-40de-ae21-a277154e3d60,af750930-5fa5-4efe-8f85-b499641377d4,f3729e05-61ef-4306-9491-6f5659d4eef9,126236de-9e7a-4814-b5d7-98cbe89b33b2,9bb31d40-c155-4213-868d-103fa0cc7c56,97bec88b-6ef7-4f24-84d6-a2bcc592548b,779b5d14-1a6f-4d0a-8763-e0a25961a811,62b4d946-e447-4fb4-96ed-b6d36c8e2c62,d475913b-0b18-47e9-90b2-28449e1d6541,77dff7ec-2e02-4b93-86d6-5d7765b8dd3d,3a0eec54-fd86-47c1-883b-d8e828d2a327,b947fc6f-1d50-4e3f-b679-8484529b787b,e8e17de2-9b5f-4f20-a18e-fdd14a8b7192,b5b6d817-c524-4bef-badc-f93876657ea2,9ab66439-8090-4f6c-ba74-874591dfd1a6,8e447490-23cd-4d64-9e3a-b8ad46c76573,762d9eca-4835-4447-a649-f91740c74bd0,5c1cf475-6fff-42e2-8508-7e1a14ea9f65,06fb659f-a4aa-4a97-a831-ed2cd2aa7e69,608770a0-440d-4349-9a1c-863e9f4d3e24,a58987e9-8b59-461a-a68b-20efa1d37d51,2d878243-5d98-489a-b8f2-8809f813b6cf,452a9534-b4fc-4483-94e0-d34b06b8e299,e690c779-3cb8-4271-8541-a699bb1bd475,d984a2a8-1209-42bb-8f59-001421c96f8f,5caa2d1f-3ab9-4e25-97a2-af29c2dbbcfb,7b1d2983-ab41-48f2-bdcd-454cbf3f81b8,b267896f-cfc0-4bbe-9fda-fdfa287d1f43,ff514fc4-bd34-4a3f-be3f-840b0e1f8091,a131d5e7-2a27-426f-90f0-25eae3c4b490,620b9bf1-f0f3-4d68-aa74-1560fb406a68,9b4c14e9-c404-4646-87b7-d3badae8a68b,9385d26b-bf44-4ff7-9410-e181413540dd,a57a4b5e-0f12-4103-8852-72bd2b7dba24,cf8040e2-39e3-4200-ab7f-b9b532310ac3,4fc7b71f-9591-4ae1-bfca-2929c7225182,da05b57a-d776-45b6-96af-1750e0254eec,ef446d15-d019-4c5a-a052-ae8cc3ea3b81,0fe4ea94-633d-4f73-985b-cbcd9f5a0270,cbb0ca6e-aa5e-4683-b107-d7dfa4cb60ce,ae6ec2ce-5c23-4e65-802f-cef9f1624d71,cd3dbc9f-a39c-4d13-af60-cba7dbc3baad,02423ab9-f3ed-4096-a7df-bfa3ebb40c7c,084b0c27-79c5-42f6-9cbd-690d0e79167a,42465540-8af1-423b-9b35-a31de1e1d49e,74a50705-7db2-44be-84dc-92b2d960d344,3b79284e-1143-46b4-b6a0-78e5448ff773,afbfc79d-6455-40b8-b475-1bcde4ef0cc9,6d705f80-8259-4ca1-8645-163f42c85a3c,129057dd-4bf4-472d-a8cf-a063dbed60ad,2833344d-bff9-4bbe-8fab-6b13d64c6b2a,48669cdc-1a64-4c99-a6e8-9a7181341fa2,be70739f-234a-4a0f-9b14-d15b8800577b,4047884e-4629-457c-9d77-130e937fae60,eb97bae4-430e-45cd-a065-2df3ab5c600e,3bcc3abf-5e76-4d3d-b0bb-c74cd0e5f4d8,c3eeecfc-de96-480f-929d-ac4a5e5a541c,3b9140da-0643-40b9-a610-fe7657942509,e6866a00-ab62-4e5d-9953-f0178ea3e570,9b731b02-db03-484d-a06e-1b55a4df57b8" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.4236937487216529, + "uncertainty": { + "num_samples": 120 + }, + "details": { + "total_matched_trajectories": "24", + "stopping_reason_count_repetition_guard": "24", + "avg_total_tokens_target_model": "14460.25", + "total_total_tokens_target_model": "347046.0", + "avg_total_tokens_other_models": "59433.04", + "total_total_tokens_other_models": "1426393.0", + "avg_total_tokens_all_models": "73893.29", + "total_total_tokens_all_models": "1773439.0", + "avg_turn_count": "26.21", + "total_turn_count": "629.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/6b177a5b-58a7-4ac2-b277-b67df1d5604f_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "3f533593ceb009f304f0601188ca2a495b5812dc705e6dca8176ff2db38e7f05", + "total_rows": 120 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/6b177a5b-58a7-4ac2-b277-b67df1d5604f_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/6b177a5b-58a7-4ac2-b277-b67df1d5604f_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..177731a23a2ef7122c7b32c3b8456aea3568f1d4 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/6b177a5b-58a7-4ac2-b277-b67df1d5604f_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f533593ceb009f304f0601188ca2a495b5812dc705e6dca8176ff2db38e7f05 +size 7716382 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/894b6600-9c98-4a0e-8376-77ef1f365c4d.json b/data/healthbench/openai/gpt-5.4-2026-03-05/894b6600-9c98-4a0e-8376-77ef1f365c4d.json new file mode 100644 index 0000000000000000000000000000000000000000..3f6c74cd35ad504e3bc5490662442870d52cfda2 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/894b6600-9c98-4a0e-8376-77ef1f365c4d.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786111974.919064", + "evaluation_timestamp": "1775604415.0", + "retrieved_timestamp": "1786111974.919064", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.202,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6189591137048764, + "uncertainty": { + "standard_deviation": 0.11205785678908674, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "49654.00", + "total_total_tokens_target_model": "99308.0", + "avg_total_tokens_other_models": "101726.50", + "total_total_tokens_other_models": "203453.0", + "avg_total_tokens_all_models": "151380.50", + "total_total_tokens_all_models": "302761.0", + "avg_turn_count": "23.00", + "total_turn_count": "46.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.11205785678908674, + "uncertainty": { + "standard_deviation": 0.11205785678908674, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "49654.00", + "total_total_tokens_target_model": "99308.0", + "avg_total_tokens_other_models": "101726.50", + "total_total_tokens_other_models": "203453.0", + "avg_total_tokens_all_models": "151380.50", + "total_total_tokens_all_models": "302761.0", + "avg_turn_count": "23.00", + "total_turn_count": "46.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.578125, + "uncertainty": { + "standard_deviation": 0.11205785678908674, + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "49654.00", + "total_total_tokens_target_model": "99308.0", + "avg_total_tokens_other_models": "101726.50", + "total_total_tokens_other_models": "203453.0", + "avg_total_tokens_all_models": "151380.50", + "total_total_tokens_all_models": "302761.0", + "avg_turn_count": "23.00", + "total_turn_count": "46.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "5", + "sample_ids": "1f548d5b-cd00-49a0-b327-283a2e00debd,0b8f1d60-2081-4562-98f7-b6a976fe1c6d,6f7a2ee9-e9c6-42d8-b79f-22dea966b8d2,19ec4833-86e9-4166-8b82-d1da09f31fd7,7ebc830a-8dbd-489b-9d61-4d8bacf0db8d" + } + }, + "evaluation_timestamp": "1775604415.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6189591137048764, + "uncertainty": { + "num_samples": 10 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "49654.00", + "total_total_tokens_target_model": "99308.0", + "avg_total_tokens_other_models": "101726.50", + "total_total_tokens_other_models": "203453.0", + "avg_total_tokens_all_models": "151380.50", + "total_total_tokens_all_models": "302761.0", + "avg_turn_count": "23.00", + "total_turn_count": "46.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/894b6600-9c98-4a0e-8376-77ef1f365c4d_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "7985697e5121b16c9273564731f2e614445f5465b808243f9f6bf256d239c2d7", + "total_rows": 10 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/894b6600-9c98-4a0e-8376-77ef1f365c4d_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/894b6600-9c98-4a0e-8376-77ef1f365c4d_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..e7b08820b566e0ec9e90b35fb01e678e2605b299 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/894b6600-9c98-4a0e-8376-77ef1f365c4d_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7985697e5121b16c9273564731f2e614445f5465b808243f9f6bf256d239c2d7 +size 745024 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/c52de382-5013-4842-a086-ae882340a976.json b/data/healthbench/openai/gpt-5.4-2026-03-05/c52de382-5013-4842-a086-ae882340a976.json new file mode 100644 index 0000000000000000000000000000000000000000..a56f1cc854dee0fb58edc2f6b187ac39b936c637 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/c52de382-5013-4842-a086-ae882340a976.json @@ -0,0 +1,322 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786117915.034618", + "evaluation_timestamp": "1776787601.0", + "retrieved_timestamp": "1786117915.034618", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+2ep/30be3326 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "3", + "sample_ids": "48c89cd1-0ecf-4738-8fcc-392269585bd3,36c6e780-d4a8-47b2-8a36-90ee4b97607f,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776787601.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0, + "uncertainty": { + "standard_deviation": 0.08676544220775459, + "num_samples": 6 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "5", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "13374.50", + "total_total_tokens_target_model": "80247.0", + "avg_total_tokens_other_models": "58397.83", + "total_total_tokens_other_models": "350387.0", + "avg_total_tokens_all_models": "71772.33", + "total_total_tokens_all_models": "430634.0", + "avg_turn_count": "33.50", + "total_turn_count": "201.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+2ep/30be3326 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "3", + "sample_ids": "48c89cd1-0ecf-4738-8fcc-392269585bd3,36c6e780-d4a8-47b2-8a36-90ee4b97607f,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776787601.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.08676544220775459, + "uncertainty": { + "standard_deviation": 0.08676544220775459, + "num_samples": 6 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "5", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "13374.50", + "total_total_tokens_target_model": "80247.0", + "avg_total_tokens_other_models": "58397.83", + "total_total_tokens_other_models": "350387.0", + "avg_total_tokens_all_models": "71772.33", + "total_total_tokens_all_models": "430634.0", + "avg_turn_count": "33.50", + "total_turn_count": "201.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+2ep/30be3326 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "3", + "sample_ids": "48c89cd1-0ecf-4738-8fcc-392269585bd3,36c6e780-d4a8-47b2-8a36-90ee4b97607f,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776787601.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.375, + "uncertainty": { + "standard_deviation": 0.08676544220775459, + "num_samples": 6 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "5", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "13374.50", + "total_total_tokens_target_model": "80247.0", + "avg_total_tokens_other_models": "58397.83", + "total_total_tokens_other_models": "350387.0", + "avg_total_tokens_all_models": "71772.33", + "total_total_tokens_all_models": "430634.0", + "avg_turn_count": "33.50", + "total_turn_count": "201.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+2ep/30be3326 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "3", + "sample_ids": "48c89cd1-0ecf-4738-8fcc-392269585bd3,36c6e780-d4a8-47b2-8a36-90ee4b97607f,886864a6-a0e4-468f-aaf4-26cf8dc5e541" + } + }, + "evaluation_timestamp": "1776787601.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": -0.012737667863316057, + "uncertainty": { + "num_samples": 6 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "5", + "stopping_reason_count_completed_with_submit": "1", + "avg_total_tokens_target_model": "13374.50", + "total_total_tokens_target_model": "80247.0", + "avg_total_tokens_other_models": "58397.83", + "total_total_tokens_other_models": "350387.0", + "avg_total_tokens_all_models": "71772.33", + "total_total_tokens_all_models": "430634.0", + "avg_turn_count": "33.50", + "total_turn_count": "201.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/c52de382-5013-4842-a086-ae882340a976_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "c27da9cb3f69a0b4c87adc38e87288b83bc6cbb73905e041cfae8b9b695c5648", + "total_rows": 6 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/c52de382-5013-4842-a086-ae882340a976_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/c52de382-5013-4842-a086-ae882340a976_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a7692091aef0adbf68f5fb73ba84ee9900da4a2c --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/c52de382-5013-4842-a086-ae882340a976_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c27da9cb3f69a0b4c87adc38e87288b83bc6cbb73905e041cfae8b9b695c5648 +size 440607 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/caf9a375-1c33-4c74-807f-052e654e4df7.json b/data/healthbench/openai/gpt-5.4-2026-03-05/caf9a375-1c33-4c74-807f-052e654e4df7.json new file mode 100644 index 0000000000000000000000000000000000000000..ae41f74b6f30a646b771a9155ba023f7d90d1ff3 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/caf9a375-1c33-4c74-807f-052e654e4df7.json @@ -0,0 +1,302 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786117990.994861", + "evaluation_timestamp": "1775733996.0", + "retrieved_timestamp": "1786117990.994861", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.49046006553034127, + "uncertainty": { + "standard_deviation": 0.045833957364667964, + "num_samples": 259 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "7847.40", + "total_total_tokens_target_model": "156948.0", + "avg_total_tokens_other_models": "61727.85", + "total_total_tokens_other_models": "1234557.0", + "avg_total_tokens_all_models": "69575.25", + "total_total_tokens_all_models": "1391505.0", + "avg_turn_count": "31.60", + "total_turn_count": "632.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.045833957364667964, + "uncertainty": { + "standard_deviation": 0.045833957364667964, + "num_samples": 259 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "7847.40", + "total_total_tokens_target_model": "156948.0", + "avg_total_tokens_other_models": "61727.85", + "total_total_tokens_other_models": "1234557.0", + "avg_total_tokens_all_models": "69575.25", + "total_total_tokens_all_models": "1391505.0", + "avg_turn_count": "31.60", + "total_turn_count": "632.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+7ep for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.435997171145686, + "uncertainty": { + "standard_deviation": 0.045833957364667964, + "num_samples": 259 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "7847.40", + "total_total_tokens_target_model": "156948.0", + "avg_total_tokens_other_models": "61727.85", + "total_total_tokens_other_models": "1234557.0", + "avg_total_tokens_all_models": "69575.25", + "total_total_tokens_all_models": "1391505.0", + "avg_turn_count": "31.60", + "total_turn_count": "632.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+7ep for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "37", + "sample_ids": "c971f9d1-5f6a-464e-b282-41c8f0d82f58,5a6e4a41-3ea6-4050-a971-93433fe34877,78510411-e03b-4f93-a369-c7e72d2e4908,bbd759b4-8e4d-4db7-bc36-f2e069396c1c,8f2a65de-dea7-48e8-8adb-6194eca26c08,6d5f483c-3e86-456d-bfd5-4e28de699ae5,f05491d8-d160-4b2b-bd5b-59d757585e39,85408180-238e-4c1b-92c1-55d8ce6c3717,ecd7453e-1dc1-4e52-bb95-67e8e5788573,0e073591-b3e7-4dc8-94d0-dc5aa93ca35c,7637b987-8fe2-49d0-8372-1ebf17284d70,5b294937-13e5-424e-8bb4-5d1904a2344a,c5bf7fc3-dfb0-4b47-9bbe-222e87b952e2,83cf8f2d-2857-4f01-a283-9595d8f4ae8e,7a4548e6-38b7-48dd-9088-d7d1219f7852,e0d1b955-1150-457e-96d7-91a9481999cd,24f9a6e7-b214-4011-94c4-6502f249a621,6a97773b-c33c-4f2e-9ac3-fdd53de0f175,f32ee0c1-ea9c-4cfb-8332-bd9579d23024,b6eb69cb-b911-4c81-a7a1-357d21d27109,c0dc053b-157b-4d13-9956-213b67ff6a36,3bb4a735-ace7-4342-b7a5-0bae083e6f82,66b448bf-2e49-488d-9d0b-f9912ab3f0f8,b12d2453-fd3b-462c-b9cc-554ff16bfd3f,0a548d04-0973-4343-a3bf-07f730bed2f6,e2cd9f23-e3dd-4b67-8984-a5412dd802e5,1fb6ab91-a999-437d-8c93-c5937c0a89a3,42af0c9f-a715-45c4-ab24-c2948f6f943d,a94aad40-387e-4a37-b92f-61badd57388b,596ba714-29f7-447e-aa73-88d46ecd6e88,1233737e-2674-41d3-8519-b817aeb0b32e,c80a2a84-281d-41cc-a10b-32ee48c584f6,bf7718f2-bceb-4702-8d74-a7fc1bd7fb78,b7a82324-73ee-4996-8f71-8f455f789e90,22fe3eec-03b0-4cdf-a06a-87e71d236082,6c568ea6-86bd-4286-94ba-e3ccc8b67b21,905949d2-7a0a-4461-8f4b-257de6be6eed" + } + }, + "evaluation_timestamp": "1775733996.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.49046006553034127, + "uncertainty": { + "num_samples": 259 + }, + "details": { + "total_matched_trajectories": "20", + "stopping_reason_count_repetition_guard": "20", + "avg_total_tokens_target_model": "7847.40", + "total_total_tokens_target_model": "156948.0", + "avg_total_tokens_other_models": "61727.85", + "total_total_tokens_other_models": "1234557.0", + "avg_total_tokens_all_models": "69575.25", + "total_total_tokens_all_models": "1391505.0", + "avg_turn_count": "31.60", + "total_turn_count": "632.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/caf9a375-1c33-4c74-807f-052e654e4df7_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "7e6ca94e64cfb42d1f3d39402e80fde360846aec21ad69172d88361f71425680", + "total_rows": 259 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/caf9a375-1c33-4c74-807f-052e654e4df7_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/caf9a375-1c33-4c74-807f-052e654e4df7_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..fe54a6e2424a356d1e7f7f9704754194950593c3 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/caf9a375-1c33-4c74-807f-052e654e4df7_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e6ca94e64cfb42d1f3d39402e80fde360846aec21ad69172d88361f71425680 +size 14857407 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/de65a62a-2b78-4c4b-85bf-47497314964d.json b/data/healthbench/openai/gpt-5.4-2026-03-05/de65a62a-2b78-4c4b-85bf-47497314964d.json new file mode 100644 index 0000000000000000000000000000000000000000..c919ee6a51d575dac88904996777214a52537ee1 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/de65a62a-2b78-4c4b-85bf-47497314964d.json @@ -0,0 +1,254 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786117907.592991", + "evaluation_timestamp": "1776259833.0", + "retrieved_timestamp": "1786117907.592991", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+1ep/044cd929 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "3", + "sample_ids": "c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5985714285714286, + "uncertainty": { + "standard_deviation": 0.1731822933531767, + "num_samples": 3 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+1ep/044cd929 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "3", + "sample_ids": "c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.1731822933531767, + "uncertainty": { + "standard_deviation": 0.1731822933531767, + "num_samples": 3 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+1ep/044cd929 for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "3", + "sample_ids": "c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.625, + "uncertainty": { + "standard_deviation": 0.1731822933531767, + "num_samples": 3 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+1ep/044cd929 for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "3", + "sample_ids": "c653b6eb-f9b9-4626-b702-ff73d71e8253,cf09b209-5434-4317-aa2a-7b07537b475e,af62e0b8-bbd0-4dcc-b55c-cb4c7bbcfb2c" + } + }, + "evaluation_timestamp": "1776259833.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5985714285714286, + "uncertainty": { + "num_samples": 3 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "64000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/de65a62a-2b78-4c4b-85bf-47497314964d_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "6b8f37a5ed8fb362fd9ae87b202b9b08fba160040081ab066f185aa88bf09d50", + "total_rows": 3 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/de65a62a-2b78-4c4b-85bf-47497314964d_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/de65a62a-2b78-4c4b-85bf-47497314964d_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a6255831866e487a7c1338a5e27b4a49664db3a7 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/de65a62a-2b78-4c4b-85bf-47497314964d_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b8f37a5ed8fb362fd9ae87b202b9b08fba160040081ab066f185aa88bf09d50 +size 133243 diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/fcc6febd-137e-401c-be76-687622239a0d.json b/data/healthbench/openai/gpt-5.4-2026-03-05/fcc6febd-137e-401c-be76-687622239a0d.json new file mode 100644 index 0000000000000000000000000000000000000000..4233194787ce3b2ecfa1192b82c843274a16f822 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/fcc6febd-137e-401c-be76-687622239a0d.json @@ -0,0 +1,318 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "healthbench/openai_gpt-5.4-2026-03-05/1786117969.209459", + "evaluation_timestamp": "1776725829.0", + "retrieved_timestamp": "1786117969.209459", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.4-2026-03-05", + "id": "openai/gpt-5.4-2026-03-05", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "bootstrap_score on healthbench/S-adaptive+C/+4ep/48e7534c for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776725829.0", + "metric_config": { + "evaluation_description": "bootstrap_score", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3384013163804875, + "uncertainty": { + "standard_deviation": 0.039908482786840116, + "num_samples": 119 + }, + "details": { + "total_matched_trajectories": "119", + "stopping_reason_count_repetition_guard": "119", + "avg_total_tokens_target_model": "12546.71", + "total_total_tokens_target_model": "1493058.0", + "avg_total_tokens_other_models": "54776.98", + "total_total_tokens_other_models": "6518461.0", + "avg_total_tokens_all_models": "67323.69", + "total_total_tokens_all_models": "8011519.0", + "avg_turn_count": "26.34", + "total_turn_count": "3135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "std on healthbench/S-adaptive+C/+4ep/48e7534c for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776725829.0", + "metric_config": { + "evaluation_description": "std", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.039908482786840116, + "uncertainty": { + "standard_deviation": 0.039908482786840116, + "num_samples": 119 + }, + "details": { + "total_matched_trajectories": "119", + "stopping_reason_count_repetition_guard": "119", + "avg_total_tokens_target_model": "12546.71", + "total_total_tokens_target_model": "1493058.0", + "avg_total_tokens_other_models": "54776.98", + "total_total_tokens_other_models": "6518461.0", + "avg_total_tokens_all_models": "67323.69", + "total_total_tokens_all_models": "8011519.0", + "avg_turn_count": "26.34", + "total_turn_count": "3135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "criteria_met_rate on healthbench/S-adaptive+C/+4ep/48e7534c for scorer _scorer", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776725829.0", + "metric_config": { + "evaluation_description": "criteria_met_rate", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3676900584795322, + "uncertainty": { + "standard_deviation": 0.039908482786840116, + "num_samples": 119 + }, + "details": { + "total_matched_trajectories": "119", + "stopping_reason_count_repetition_guard": "119", + "avg_total_tokens_target_model": "12546.71", + "total_total_tokens_target_model": "1493058.0", + "avg_total_tokens_other_models": "54776.98", + "total_total_tokens_other_models": "6518461.0", + "avg_total_tokens_all_models": "67323.69", + "total_total_tokens_all_models": "8011519.0", + "avg_turn_count": "26.34", + "total_turn_count": "3135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + }, + { + "evaluation_name": "mean on healthbench/S-adaptive+C/+4ep/48e7534c for scorer healthbench_score", + "source_data": { + "dataset_name": "healthbench", + "source_type": "other", + "additional_details": { + "shuffled": "False", + "inspect_dataset_location": "/home/ubuntu/.cache/inspect_evals/healthbench/full_c37d29dcb36dd46fd8b36341c84a8d06.jsonl", + "inspect_dataset_name": "full_c37d29dcb36dd46fd8b36341c84a8d06", + "samples_number": "30", + "sample_ids": "269074dc-9495-484b-b2f9-0f0943c0f816,b97b1bae-b09f-4073-8e16-540bba766c60,8e88f21a-24c4-4f9c-b848-faf5e10e31b9,a49eb9af-926e-4f7c-94d0-133729059116,1049130c-e9c9-461d-b080-90027bc011c0,13bb02e4-844e-4972-9200-2befa5f911d8,2694a753-2f57-41eb-a7f9-8c4c7acdd2e8,b160df9f-be98-4a4d-9cc8-11a612ffaf87,5e0a84d7-b972-484c-a649-469b15f0a90f,c7606b1b-ca4d-4a71-8277-bcddff09a653,20cdebb7-9189-4f09-ad21-bfc8e8b75680,f24935ac-0530-4063-b2f1-ac45cd265037,1b8f338a-2e98-4a42-ba80-cba266f4b9fa,6f827e79-05c4-4383-8ed4-16d36260f29f,c835bebe-e998-4dcd-843a-2de03d4aabc6,70afd1ac-85ea-4455-a5a2-15b35378e10a,aa7a760a-5173-4a39-af83-48d6ae2230f0,f712eb09-82af-4fe8-800d-bc3af772a9ba,8ee6a990-8966-4b42-baa8-6c7e9f74ce74,267bda16-fd4b-4a9e-b3ac-5602c2532eed,0e819a9c-851d-4a7a-9263-62cfc8ce1b48,328c55cf-0b57-4992-99b1-f6a21af62ade,7b415a2f-22b0-4735-a38e-567f1891acc5,c49ac752-3693-4be1-ab4d-91fbdd02f827,a079d928-7b78-4a85-a72b-4274d7fc2914,a20cbbf3-330b-46d9-a35b-92e89a022282,3a896889-9029-4ef8-8684-ca378da4702b,7074bb7a-a595-434d-883e-95ef6f5f6667,9f7b555a-fc30-40c3-88d3-f2ece06454a2,c0d80b93-b2d5-4129-b998-b1312f98e2a8" + } + }, + "evaluation_timestamp": "1776725829.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3384013163804875, + "uncertainty": { + "num_samples": 119 + }, + "details": { + "total_matched_trajectories": "119", + "stopping_reason_count_repetition_guard": "119", + "avg_total_tokens_target_model": "12546.71", + "total_total_tokens_target_model": "1493058.0", + "avg_total_tokens_other_models": "54776.98", + "total_total_tokens_other_models": "6518461.0", + "avg_total_tokens_all_models": "67323.69", + "total_total_tokens_all_models": "8011519.0", + "avg_turn_count": "26.34", + "total_turn_count": "3135.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"healthbench_score_value\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "3600", + "attempt_timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 10000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "3600", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/healthbench/openai/gpt-5.4-2026-03-05/fcc6febd-137e-401c-be76-687622239a0d_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "bcb1e74079b1ec7ee8ae43c16651940b1aa8869d273cf012382b518453040933", + "total_rows": 119 + } +} \ No newline at end of file diff --git a/data/healthbench/openai/gpt-5.4-2026-03-05/fcc6febd-137e-401c-be76-687622239a0d_samples.jsonl b/data/healthbench/openai/gpt-5.4-2026-03-05/fcc6febd-137e-401c-be76-687622239a0d_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b84587438cb80a308d521c5cff8f1afaee6d0104 --- /dev/null +++ b/data/healthbench/openai/gpt-5.4-2026-03-05/fcc6febd-137e-401c-be76-687622239a0d_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcb1e74079b1ec7ee8ae43c16651940b1aa8869d273cf012382b518453040933 +size 7810396 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/08c924b5-14fa-414c-9a10-baa175821c9a.json b/data/hle/openai/gpt-5.2-2025-12-11/08c924b5-14fa-414c-9a10-baa175821c9a.json new file mode 100644 index 0000000000000000000000000000000000000000..aff7d36def14b862f8ddc5c2de636223e8c3d917 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/08c924b5-14fa-414c-9a10-baa175821c9a.json @@ -0,0 +1,113 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118273.574258", + "evaluation_timestamp": "1777335634.0", + "retrieved_timestamp": "1786118273.574258", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/76ebb4ee for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 4, + "sample_ids": [ + "hle_ca59b32c", + "hle_e3817b7e", + "hle_35a40c52", + "hle_cd4aa5b3" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1777335634.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.25, + "uncertainty": { + "standard_error": { + "value": 0.14433756729740643 + }, + "num_samples": 8 + }, + "details": { + "total_matched_trajectories": "8", + "stopping_reason_count_repetition_guard": "6", + "stopping_reason_count_completed_on_successful_submit": "2", + "avg_total_tokens_target_model": "25483.00", + "total_total_tokens_target_model": "203864.0", + "avg_total_tokens_other_models": "4668.00", + "total_total_tokens_other_models": "37344.0", + "avg_total_tokens_all_models": "30151.00", + "total_total_tokens_all_models": "241208.0", + "avg_turn_count": "5.50", + "total_turn_count": "44.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/08c924b5-14fa-414c-9a10-baa175821c9a_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "b21da96efd1e109ab302e360bc1b926902c1884c400e0b977c16931908898e80", + "total_rows": 8 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/08c924b5-14fa-414c-9a10-baa175821c9a_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/08c924b5-14fa-414c-9a10-baa175821c9a_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..875e8a43b9ee5674c5de06dbacece49d6b7d2e51 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/08c924b5-14fa-414c-9a10-baa175821c9a_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b21da96efd1e109ab302e360bc1b926902c1884c400e0b977c16931908898e80 +size 159307 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/0a306816-cfc4-49e3-af56-7350e426dc16.json b/data/hle/openai/gpt-5.2-2025-12-11/0a306816-cfc4-49e3-af56-7350e426dc16.json new file mode 100644 index 0000000000000000000000000000000000000000..e53d66299af0a428371af9a36bedb09aba785c21 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/0a306816-cfc4-49e3-af56-7350e426dc16.json @@ -0,0 +1,111 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118296.223354", + "evaluation_timestamp": "1781949668.0", + "retrieved_timestamp": "1786118296.223354", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+3ep/724f42d8 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 2, + "sample_ids": [ + "hle_a559277e", + "hle_376b8247" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1781949668.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0, + "uncertainty": { + "num_samples": 2 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "32573.00", + "total_total_tokens_target_model": "65146.0", + "avg_total_tokens_other_models": "4756.50", + "total_total_tokens_other_models": "9513.0", + "avg_total_tokens_all_models": "37329.50", + "total_total_tokens_all_models": "74659.0", + "avg_turn_count": "5.00", + "total_turn_count": "10.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "16", + "max_tokens": "20000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/0a306816-cfc4-49e3-af56-7350e426dc16_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "54cd402bfd90d5c0a47a7db07de6f8a0ef845dd13969027c7a471ce9987b7356", + "total_rows": 2 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/0a306816-cfc4-49e3-af56-7350e426dc16_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/0a306816-cfc4-49e3-af56-7350e426dc16_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..3f42765f7ac2e96045789c02234be63b37dee3f2 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/0a306816-cfc4-49e3-af56-7350e426dc16_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:54cd402bfd90d5c0a47a7db07de6f8a0ef845dd13969027c7a471ce9987b7356 +size 42398 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/0c5e0246-1861-4e6a-85c6-42daa494c7d0.json b/data/hle/openai/gpt-5.2-2025-12-11/0c5e0246-1861-4e6a-85c6-42daa494c7d0.json new file mode 100644 index 0000000000000000000000000000000000000000..cc07186eae0a73718c398a8c6004c4a84bde6aae --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/0c5e0246-1861-4e6a-85c6-42daa494c7d0.json @@ -0,0 +1,113 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118290.501853", + "evaluation_timestamp": "1779023464.0", + "retrieved_timestamp": "1786118290.501853", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/b79c41ef for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 4, + "sample_ids": [ + "hle_28643173", + "hle_ebd62885", + "hle_34c69126", + "hle_c2ff3a43" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1779023464.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 5 + }, + "details": { + "total_matched_trajectories": "5", + "stopping_reason_count_completed_on_successful_submit": "5", + "avg_total_tokens_target_model": "20699.60", + "total_total_tokens_target_model": "103498.0", + "avg_total_tokens_other_models": "2151.20", + "total_total_tokens_other_models": "10756.0", + "avg_total_tokens_all_models": "22850.80", + "total_total_tokens_all_models": "114254.0", + "avg_turn_count": "1.80", + "total_turn_count": "9.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "8", + "max_tokens": "32000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "8", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/0c5e0246-1861-4e6a-85c6-42daa494c7d0_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "50d70a1ea3a072a679b13f77ff82a8ff1ee66a8d971a71135214c786af8b1e42", + "total_rows": 5 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/0c5e0246-1861-4e6a-85c6-42daa494c7d0_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/0c5e0246-1861-4e6a-85c6-42daa494c7d0_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..6fda77dfb17627a442477f5c81394381b1d27749 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/0c5e0246-1861-4e6a-85c6-42daa494c7d0_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50d70a1ea3a072a679b13f77ff82a8ff1ee66a8d971a71135214c786af8b1e42 +size 327879 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/1340a5a0-5c5a-4d9d-b785-cb3622226986.json b/data/hle/openai/gpt-5.2-2025-12-11/1340a5a0-5c5a-4d9d-b785-cb3622226986.json new file mode 100644 index 0000000000000000000000000000000000000000..9fe78c7d2c4af6cbb49d045cee1d76dc8cba5702 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/1340a5a0-5c5a-4d9d-b785-cb3622226986.json @@ -0,0 +1,102 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118227.265968", + "evaluation_timestamp": "1783093931.0", + "retrieved_timestamp": "1786118227.265968", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/1b4988d8 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 2, + "sample_ids": [ + "hle_a559277e", + "hle_25e0c71d" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1783093931.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5, + "uncertainty": { + "standard_error": { + "value": 0.5 + }, + "num_samples": 2 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "1", + "timeout": "1200", + "attempt_timeout": "3600", + "max_connections": "16", + "max_tokens": "64000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 1 + }, + "additional_details": { + "max_retries": "1", + "timeout": "1200", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/1340a5a0-5c5a-4d9d-b785-cb3622226986_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "083642b48f4da877316d67ffaa51bf7660cda0a4469f5bda27a45f094e547e83", + "total_rows": 2 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/1340a5a0-5c5a-4d9d-b785-cb3622226986_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/1340a5a0-5c5a-4d9d-b785-cb3622226986_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c67d9ad164688e06ba62a45b08290cc778e802c1 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/1340a5a0-5c5a-4d9d-b785-cb3622226986_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:083642b48f4da877316d67ffaa51bf7660cda0a4469f5bda27a45f094e547e83 +size 10734 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/15b88acb-bbc3-4996-a9d1-afd4b4168bf8.json b/data/hle/openai/gpt-5.2-2025-12-11/15b88acb-bbc3-4996-a9d1-afd4b4168bf8.json new file mode 100644 index 0000000000000000000000000000000000000000..9c6a03877b8f8c550fcbd1fb533d2e9b376d88b8 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/15b88acb-bbc3-4996-a9d1-afd4b4168bf8.json @@ -0,0 +1,112 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118271.415644", + "evaluation_timestamp": "1776809967.0", + "retrieved_timestamp": "1786118271.415644", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/263c9a07 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 3, + "sample_ids": [ + "hle_86a18059", + "hle_85e95ef4", + "hle_0702accc" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776809967.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3333333333333333, + "uncertainty": { + "standard_error": { + "value": 0.33333333333333337 + }, + "num_samples": 6 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_completed_on_successful_submit": "2", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "19774.67", + "total_total_tokens_target_model": "118648.0", + "avg_total_tokens_other_models": "2940.17", + "total_total_tokens_other_models": "17641.0", + "avg_total_tokens_all_models": "22714.83", + "total_total_tokens_all_models": "136289.0", + "avg_turn_count": "4.17", + "total_turn_count": "25.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/15b88acb-bbc3-4996-a9d1-afd4b4168bf8_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "ad310fd887d5997e713e31d994796c01311931e4b3565c7136931bbb76216baa", + "total_rows": 6 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/15b88acb-bbc3-4996-a9d1-afd4b4168bf8_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/15b88acb-bbc3-4996-a9d1-afd4b4168bf8_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..005aae973a48c4da628d557b5377541845e51de6 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/15b88acb-bbc3-4996-a9d1-afd4b4168bf8_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad310fd887d5997e713e31d994796c01311931e4b3565c7136931bbb76216baa +size 145454 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/17589734-de5e-49f0-8025-e1e6306ce759.json b/data/hle/openai/gpt-5.2-2025-12-11/17589734-de5e-49f0-8025-e1e6306ce759.json new file mode 100644 index 0000000000000000000000000000000000000000..d32d4594cb3cd7b2ff22100713ac73d1e4230137 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/17589734-de5e-49f0-8025-e1e6306ce759.json @@ -0,0 +1,116 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118233.194746", + "evaluation_timestamp": "1782219554.0", + "retrieved_timestamp": "1786118233.194746", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/9c119114 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 3, + "sample_ids": [ + "hle_376b8247", + "hle_70e5aebc", + "hle_34c69126" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1782219554.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6666666666666666, + "uncertainty": { + "standard_error": { + "value": 0.33333333333333337 + }, + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_completed_on_successful_submit": "2", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "70047.33", + "total_total_tokens_target_model": "210142.0", + "avg_total_tokens_other_models": "2727.33", + "total_total_tokens_other_models": "8182.0", + "avg_total_tokens_all_models": "72774.67", + "total_total_tokens_all_models": "218324.0", + "avg_turn_count": "4.00", + "total_turn_count": "12.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "1200", + "max_connections": "16", + "max_tokens": "20000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/17589734-de5e-49f0-8025-e1e6306ce759_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "6681e16f5cd0df794627d679f40a245d88930928101fe617256068dcabdb0515", + "total_rows": 3 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/17589734-de5e-49f0-8025-e1e6306ce759_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/17589734-de5e-49f0-8025-e1e6306ce759_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..661ec8bd3e6ac5c8e29f062122647d1d8811e278 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/17589734-de5e-49f0-8025-e1e6306ce759_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6681e16f5cd0df794627d679f40a245d88930928101fe617256068dcabdb0515 +size 44827 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/17e1cdaa-38bf-4b4f-9008-bd5be8f3ef87.json b/data/hle/openai/gpt-5.2-2025-12-11/17e1cdaa-38bf-4b4f-9008-bd5be8f3ef87.json new file mode 100644 index 0000000000000000000000000000000000000000..cb8874c7fe31a8c3ed5ed9108f16f84ec818584a --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/17e1cdaa-38bf-4b4f-9008-bd5be8f3ef87.json @@ -0,0 +1,107 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118294.09439", + "evaluation_timestamp": "1777041696.0", + "retrieved_timestamp": "1786118294.09439", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+3ep/d70965ab for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 2, + "sample_ids": [ + "hle_9282e7ae", + "hle_aa2fed90" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1777041696.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 6 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_completed_on_successful_submit": "6", + "avg_total_tokens_target_model": "9481.17", + "total_total_tokens_target_model": "56887.0", + "avg_total_tokens_other_models": "1890.67", + "total_total_tokens_other_models": "11344.0", + "avg_total_tokens_all_models": "11371.83", + "total_total_tokens_all_models": "68231.0", + "avg_turn_count": "1.67", + "total_turn_count": "10.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/17e1cdaa-38bf-4b4f-9008-bd5be8f3ef87_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "1feb5bbd137d5bb75ad183d3b649797282ccaba8d88db3dcc1c88d26516c7f96", + "total_rows": 6 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/17e1cdaa-38bf-4b4f-9008-bd5be8f3ef87_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/17e1cdaa-38bf-4b4f-9008-bd5be8f3ef87_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..f35d7c4327bc659f7fd78ee89638a22a92f0c1bd --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/17e1cdaa-38bf-4b4f-9008-bd5be8f3ef87_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1feb5bbd137d5bb75ad183d3b649797282ccaba8d88db3dcc1c88d26516c7f96 +size 247850 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/481c53cb-5bf6-4595-b02d-36c011c7f6f8.json b/data/hle/openai/gpt-5.2-2025-12-11/481c53cb-5bf6-4595-b02d-36c011c7f6f8.json new file mode 100644 index 0000000000000000000000000000000000000000..d24579b5a799f634ab7f8f3ec028388b9f1f089b --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/481c53cb-5bf6-4595-b02d-36c011c7f6f8.json @@ -0,0 +1,115 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118298.719428", + "evaluation_timestamp": "1782228687.0", + "retrieved_timestamp": "1786118298.719428", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+3ep/1b4988d8 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 2, + "sample_ids": [ + "hle_a559277e", + "hle_25e0c71d" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1782228687.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.6666666666666666, + "uncertainty": { + "standard_error": { + "value": 0.3333333333333333 + }, + "num_samples": 6 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_completed_on_successful_submit": "4", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "212576.17", + "total_total_tokens_target_model": "1275457.0", + "avg_total_tokens_other_models": "2213.50", + "total_total_tokens_other_models": "13281.0", + "avg_total_tokens_all_models": "214789.67", + "total_total_tokens_all_models": "1288738.0", + "avg_turn_count": "10.83", + "total_turn_count": "65.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "1200", + "max_connections": "16", + "max_tokens": "20000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/481c53cb-5bf6-4595-b02d-36c011c7f6f8_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "aaa7cd870c6da817d217fe6f27ee35446e8fd6a5b526b5144255cb0bc7993923", + "total_rows": 6 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/481c53cb-5bf6-4595-b02d-36c011c7f6f8_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/481c53cb-5bf6-4595-b02d-36c011c7f6f8_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a4372f902df52302c8ac67f0cc49f46c550f7b54 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/481c53cb-5bf6-4595-b02d-36c011c7f6f8_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aaa7cd870c6da817d217fe6f27ee35446e8fd6a5b526b5144255cb0bc7993923 +size 143335 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/5290ba3e-d404-4ce7-a4d8-b794a6825171.json b/data/hle/openai/gpt-5.2-2025-12-11/5290ba3e-d404-4ce7-a4d8-b794a6825171.json new file mode 100644 index 0000000000000000000000000000000000000000..0259be49c30cf0e15bd12f304d9539727d685396 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/5290ba3e-d404-4ce7-a4d8-b794a6825171.json @@ -0,0 +1,111 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118309.587027", + "evaluation_timestamp": "1776907849.0", + "retrieved_timestamp": "1786118309.587027", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+3ep/676471b5 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 6, + "sample_ids": [ + "hle_3cf4483d", + "hle_f85a0fa1", + "hle_31497469", + "hle_d0c8b07e", + "hle_cf3d65bd", + "hle_33bc94e2" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776907849.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 16 + }, + "details": { + "total_matched_trajectories": "16", + "stopping_reason_count_completed_on_successful_submit": "16", + "avg_total_tokens_target_model": "8711.25", + "total_total_tokens_target_model": "139380.0", + "avg_total_tokens_other_models": "1531.06", + "total_total_tokens_other_models": "24497.0", + "avg_total_tokens_all_models": "10242.31", + "total_total_tokens_all_models": "163877.0", + "avg_turn_count": "1.50", + "total_turn_count": "24.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/5290ba3e-d404-4ce7-a4d8-b794a6825171_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "4e4323a0bab3161845b9adba584723f1a4163fbb3e3151c71fd8b6e5dfd119af", + "total_rows": 16 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/5290ba3e-d404-4ce7-a4d8-b794a6825171_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/5290ba3e-d404-4ce7-a4d8-b794a6825171_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c833116b90b7d7fc24a80833005bd530343f4ef8 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/5290ba3e-d404-4ce7-a4d8-b794a6825171_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e4323a0bab3161845b9adba584723f1a4163fbb3e3151c71fd8b6e5dfd119af +size 531283 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/63c615d0-82e8-4a67-82ec-80668884a9ec.json b/data/hle/openai/gpt-5.2-2025-12-11/63c615d0-82e8-4a67-82ec-80668884a9ec.json new file mode 100644 index 0000000000000000000000000000000000000000..52c17e58ebc32bdc27daa77078297df724f28410 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/63c615d0-82e8-4a67-82ec-80668884a9ec.json @@ -0,0 +1,111 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118236.791224", + "evaluation_timestamp": "1776519735.0", + "retrieved_timestamp": "1786118236.791224", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/0af7e9fe for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 4, + "sample_ids": [ + "hle_f46c2857", + "hle_d1bd2767", + "hle_e1296346", + "hle_8a419425" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776519735.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.25, + "uncertainty": { + "standard_error": { + "value": 0.25 + }, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_on_successful_submit": "1", + "stopping_reason_count_repetition_guard": "3", + "avg_total_tokens_target_model": "43517.00", + "total_total_tokens_target_model": "174068.0", + "avg_total_tokens_other_models": "4650.50", + "total_total_tokens_other_models": "18602.0", + "avg_total_tokens_all_models": "48167.50", + "total_total_tokens_all_models": "192670.0", + "avg_turn_count": "4.50", + "total_turn_count": "18.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/63c615d0-82e8-4a67-82ec-80668884a9ec_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "b9d8346f6862f9c82554d8712ef6c07b51461a533b581982606ca7ca7a76c823", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/63c615d0-82e8-4a67-82ec-80668884a9ec_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/63c615d0-82e8-4a67-82ec-80668884a9ec_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..db3ed4ca2e597622468bb2657dcca846d52914e8 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/63c615d0-82e8-4a67-82ec-80668884a9ec_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9d8346f6862f9c82554d8712ef6c07b51461a533b581982606ca7ca7a76c823 +size 76022 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/6d363ed8-a673-49c6-86b6-0b2d79984dbf.json b/data/hle/openai/gpt-5.2-2025-12-11/6d363ed8-a673-49c6-86b6-0b2d79984dbf.json new file mode 100644 index 0000000000000000000000000000000000000000..2824c09f1ed109a35f4d4ea13a009bdfdf5275d1 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/6d363ed8-a673-49c6-86b6-0b2d79984dbf.json @@ -0,0 +1,106 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118224.111437", + "evaluation_timestamp": "1777025746.0", + "retrieved_timestamp": "1786118224.111437", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/537cc44d for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 1, + "sample_ids": [ + "hle_f85a0fa1" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1777025746.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 1 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "2692.00", + "total_total_tokens_target_model": "2692.0", + "avg_total_tokens_other_models": "721.00", + "total_total_tokens_other_models": "721.0", + "avg_total_tokens_all_models": "3413.00", + "total_total_tokens_all_models": "3413.0", + "avg_turn_count": "1.00", + "total_turn_count": "1.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/6d363ed8-a673-49c6-86b6-0b2d79984dbf_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "6cf2e16e9958fa47a9686cce63b02ea9e63e8abcee98b59981f8c36b92d80f4e", + "total_rows": 1 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/6d363ed8-a673-49c6-86b6-0b2d79984dbf_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/6d363ed8-a673-49c6-86b6-0b2d79984dbf_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..5d0d3daa3caa577d1f66e03c1c395ea96e311449 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/6d363ed8-a673-49c6-86b6-0b2d79984dbf_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6cf2e16e9958fa47a9686cce63b02ea9e63e8abcee98b59981f8c36b92d80f4e +size 15212 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/7987129d-63d3-44cc-b299-bed29b5c3656.json b/data/hle/openai/gpt-5.2-2025-12-11/7987129d-63d3-44cc-b299-bed29b5c3656.json new file mode 100644 index 0000000000000000000000000000000000000000..31cf184f4c256bab4eab02da417cb83d48c7f189 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/7987129d-63d3-44cc-b299-bed29b5c3656.json @@ -0,0 +1,102 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118300.907145", + "evaluation_timestamp": "1782771326.0", + "retrieved_timestamp": "1786118300.907145", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+3ep/1b4988d8 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 2, + "sample_ids": [ + "hle_a559277e", + "hle_25e0c71d" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1782771326.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5, + "uncertainty": { + "standard_error": { + "value": 0.5 + }, + "num_samples": 4 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "1200", + "max_connections": "16", + "max_tokens": "20000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/7987129d-63d3-44cc-b299-bed29b5c3656_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "8104522cddf43aaeed856bf42a2e877678fbac70ba19ca5563f6a9cf2e95aecd", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/7987129d-63d3-44cc-b299-bed29b5c3656_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/7987129d-63d3-44cc-b299-bed29b5c3656_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b7ff4b7e019c7d2c1092e927a618fd1ff337e731 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/7987129d-63d3-44cc-b299-bed29b5c3656_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8104522cddf43aaeed856bf42a2e877678fbac70ba19ca5563f6a9cf2e95aecd +size 140471 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/84bdbf55-e911-43e6-9714-6bc240da05e0.json b/data/hle/openai/gpt-5.2-2025-12-11/84bdbf55-e911-43e6-9714-6bc240da05e0.json new file mode 100644 index 0000000000000000000000000000000000000000..948344b4d7aafd0fdb34834fb25d39cfe706a0be --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/84bdbf55-e911-43e6-9714-6bc240da05e0.json @@ -0,0 +1,114 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118260.402608", + "evaluation_timestamp": "1777369892.0", + "retrieved_timestamp": "1786118260.402608", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/30fc061f for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 5, + "sample_ids": [ + "hle_4418bea4", + "hle_9019c023", + "hle_1dc141bb", + "hle_fcd42d0c", + "hle_c6adc1b4" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1777369892.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.25, + "uncertainty": { + "standard_error": { + "value": 0.25 + }, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_on_successful_submit": "1", + "stopping_reason_count_repetition_guard": "3", + "avg_total_tokens_target_model": "28967.25", + "total_total_tokens_target_model": "115869.0", + "avg_total_tokens_other_models": "4274.00", + "total_total_tokens_other_models": "17096.0", + "avg_total_tokens_all_models": "33241.25", + "total_total_tokens_all_models": "132965.0", + "avg_turn_count": "6.25", + "total_turn_count": "25.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/84bdbf55-e911-43e6-9714-6bc240da05e0_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "cf3b25e1d4bc019c909972cd178b467ed8e1f869c07607da1b3840374f04e6b3", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/84bdbf55-e911-43e6-9714-6bc240da05e0_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/84bdbf55-e911-43e6-9714-6bc240da05e0_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..68f32f55ce5be5d2cf1e0388d7e171d0a2e81f4b --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/84bdbf55-e911-43e6-9714-6bc240da05e0_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf3b25e1d4bc019c909972cd178b467ed8e1f869c07607da1b3840374f04e6b3 +size 157232 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/9cf0c797-27fc-41eb-8943-d71507d94bca.json b/data/hle/openai/gpt-5.2-2025-12-11/9cf0c797-27fc-41eb-8943-d71507d94bca.json new file mode 100644 index 0000000000000000000000000000000000000000..83217f478e8be2f41032449d065ab0ccf6797b3f --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/9cf0c797-27fc-41eb-8943-d71507d94bca.json @@ -0,0 +1,114 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118292.286745", + "evaluation_timestamp": "1778936860.0", + "retrieved_timestamp": "1786118292.286745", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/497bd21e for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 5, + "sample_ids": [ + "hle_28643173", + "hle_ebd62885", + "hle_70e5aebc", + "hle_34c69126", + "hle_c2ff3a43" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1778936860.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 1 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "30730.00", + "total_total_tokens_target_model": "30730.0", + "avg_total_tokens_other_models": "1514.00", + "total_total_tokens_other_models": "1514.0", + "avg_total_tokens_all_models": "32244.00", + "total_total_tokens_all_models": "32244.0", + "avg_turn_count": "3.00", + "total_turn_count": "3.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "8", + "max_tokens": "32000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "8", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/9cf0c797-27fc-41eb-8943-d71507d94bca_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e6d233fb94d9aa2cbef6f7a10e2cb5ff189bf19da50ec12f25785c2662a5b3ca", + "total_rows": 1 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/9cf0c797-27fc-41eb-8943-d71507d94bca_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/9cf0c797-27fc-41eb-8943-d71507d94bca_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..54496b1b451327ca16fc321a19070fc33875b0a5 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/9cf0c797-27fc-41eb-8943-d71507d94bca_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6d233fb94d9aa2cbef6f7a10e2cb5ff189bf19da50ec12f25785c2662a5b3ca +size 12511 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/9d0ad869-666f-4d0b-86c6-6134692c978e.json b/data/hle/openai/gpt-5.2-2025-12-11/9d0ad869-666f-4d0b-86c6-6134692c978e.json new file mode 100644 index 0000000000000000000000000000000000000000..9bb880d2a516b731d9b99fcaf2f107afc03e202b --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/9d0ad869-666f-4d0b-86c6-6134692c978e.json @@ -0,0 +1,110 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118225.719258", + "evaluation_timestamp": "1778931174.0", + "retrieved_timestamp": "1786118225.719258", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/f59fd081 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 1, + "sample_ids": [ + "hle_d7b7c3dc" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1778931174.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 1 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "20933.00", + "total_total_tokens_target_model": "20933.0", + "avg_total_tokens_other_models": "1008.00", + "total_total_tokens_other_models": "1008.0", + "avg_total_tokens_all_models": "21941.00", + "total_total_tokens_all_models": "21941.0", + "avg_turn_count": "1.00", + "total_turn_count": "1.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "8", + "max_tokens": "32000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "8", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/9d0ad869-666f-4d0b-86c6-6134692c978e_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "c57b10b51822c847f8bc04608b5e69e88884deabdcb3d688a9c2fc5dfda7bc3c", + "total_rows": 1 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/9d0ad869-666f-4d0b-86c6-6134692c978e_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/9d0ad869-666f-4d0b-86c6-6134692c978e_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..4db51ac7d229e94127189ba6ed16e8001c95a89c --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/9d0ad869-666f-4d0b-86c6-6134692c978e_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c57b10b51822c847f8bc04608b5e69e88884deabdcb3d688a9c2fc5dfda7bc3c +size 105101 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/9f401cfb-567a-4b8f-8708-41fd99ceb8bf.json b/data/hle/openai/gpt-5.2-2025-12-11/9f401cfb-567a-4b8f-8708-41fd99ceb8bf.json new file mode 100644 index 0000000000000000000000000000000000000000..bce46f97e60f5bb677d30b66954e093e45a9fd02 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/9f401cfb-567a-4b8f-8708-41fd99ceb8bf.json @@ -0,0 +1,110 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118265.842828", + "evaluation_timestamp": "1782186032.0", + "retrieved_timestamp": "1786118265.842828", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/57058abc for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 1, + "sample_ids": [ + "hle_c2ff3a43" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1782186032.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 1 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "44981.00", + "total_total_tokens_target_model": "44981.0", + "avg_total_tokens_other_models": "2630.00", + "total_total_tokens_other_models": "2630.0", + "avg_total_tokens_all_models": "47611.00", + "total_total_tokens_all_models": "47611.0", + "avg_turn_count": "5.00", + "total_turn_count": "5.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "16", + "max_tokens": "20000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/9f401cfb-567a-4b8f-8708-41fd99ceb8bf_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "50d2226021d69a23287555b05494b84d87907c881390bfb7dbccf959d0fb4b39", + "total_rows": 1 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/9f401cfb-567a-4b8f-8708-41fd99ceb8bf_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/9f401cfb-567a-4b8f-8708-41fd99ceb8bf_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..d30b5a7ed48a49d036d5dca47cd4365f8f9c9ffc --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/9f401cfb-567a-4b8f-8708-41fd99ceb8bf_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50d2226021d69a23287555b05494b84d87907c881390bfb7dbccf959d0fb4b39 +size 17433 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/a41a8781-e1cc-4e30-92a1-59083db08939.json b/data/hle/openai/gpt-5.2-2025-12-11/a41a8781-e1cc-4e30-92a1-59083db08939.json new file mode 100644 index 0000000000000000000000000000000000000000..5195d92165bcbe0deb243f45e3cc71d93110a63c --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/a41a8781-e1cc-4e30-92a1-59083db08939.json @@ -0,0 +1,106 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118262.090966", + "evaluation_timestamp": "1777386701.0", + "retrieved_timestamp": "1786118262.090966", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/d1f7f290 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 1, + "sample_ids": [ + "hle_0a1eec3f" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1777386701.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 2 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_completed_on_successful_submit": "2", + "avg_total_tokens_target_model": "14094.50", + "total_total_tokens_target_model": "28189.0", + "avg_total_tokens_other_models": "1444.00", + "total_total_tokens_other_models": "2888.0", + "avg_total_tokens_all_models": "15538.50", + "total_total_tokens_all_models": "31077.0", + "avg_turn_count": "2.00", + "total_turn_count": "4.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/a41a8781-e1cc-4e30-92a1-59083db08939_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "c4b9a62e878f0d6a4865ced3decf6ffbbfd0b9fdbea526b01d054ccd8355efab", + "total_rows": 2 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/a41a8781-e1cc-4e30-92a1-59083db08939_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/a41a8781-e1cc-4e30-92a1-59083db08939_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..cf5f171b8385ec90a0ecb0370e1d7578abd44e55 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/a41a8781-e1cc-4e30-92a1-59083db08939_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4b9a62e878f0d6a4865ced3decf6ffbbfd0b9fdbea526b01d054ccd8355efab +size 76024 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/a9e5dd04-0445-437f-aa28-2ae74930be9b.json b/data/hle/openai/gpt-5.2-2025-12-11/a9e5dd04-0445-437f-aa28-2ae74930be9b.json new file mode 100644 index 0000000000000000000000000000000000000000..5efec899d93b716ac464c0b74af276e24b423021 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/a9e5dd04-0445-437f-aa28-2ae74930be9b.json @@ -0,0 +1,109 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118256.944824", + "evaluation_timestamp": "1776803720.0", + "retrieved_timestamp": "1786118256.944824", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/903a2cff for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 4, + "sample_ids": [ + "hle_76cdecee", + "hle_142ffcff", + "hle_ffb12d83", + "hle_37c97863" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776803720.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 1.0, + "uncertainty": { + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_on_successful_submit": "4", + "avg_total_tokens_target_model": "14472.50", + "total_total_tokens_target_model": "57890.0", + "avg_total_tokens_other_models": "1645.00", + "total_total_tokens_other_models": "6580.0", + "avg_total_tokens_all_models": "16117.50", + "total_total_tokens_all_models": "64470.0", + "avg_turn_count": "4.00", + "total_turn_count": "16.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/a9e5dd04-0445-437f-aa28-2ae74930be9b_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e26dc9f485ebff3b42f5edb5c6a822ea78fedb426230a8f4c08ed95aed0aff96", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/a9e5dd04-0445-437f-aa28-2ae74930be9b_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/a9e5dd04-0445-437f-aa28-2ae74930be9b_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..66530f63aa085db2c40532cc93b54e42cb5ce672 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/a9e5dd04-0445-437f-aa28-2ae74930be9b_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e26dc9f485ebff3b42f5edb5c6a822ea78fedb426230a8f4c08ed95aed0aff96 +size 114900 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/b46321ef-c664-4379-878a-00c030df0f95.json b/data/hle/openai/gpt-5.2-2025-12-11/b46321ef-c664-4379-878a-00c030df0f95.json new file mode 100644 index 0000000000000000000000000000000000000000..a29536fccd0b41d9f54dcd75b3899ab16f212557 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/b46321ef-c664-4379-878a-00c030df0f95.json @@ -0,0 +1,116 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118234.895412", + "evaluation_timestamp": "1783327709.0", + "retrieved_timestamp": "1786118234.895412", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/837a57e8 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 3, + "sample_ids": [ + "hle_dfdb78f5", + "hle_e3b8f5c2", + "hle_45d1c140" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1783327709.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5, + "uncertainty": { + "standard_error": { + "value": 0.5 + }, + "num_samples": 2 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_repetition_guard": "1", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "554605.50", + "total_total_tokens_target_model": "1109211.0", + "avg_total_tokens_other_models": "6215.50", + "total_total_tokens_other_models": "12431.0", + "avg_total_tokens_all_models": "560821.00", + "total_total_tokens_all_models": "1121642.0", + "avg_turn_count": "9.50", + "total_turn_count": "19.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "1", + "timeout": "1200", + "attempt_timeout": "3600", + "max_connections": "16", + "max_tokens": "64000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 1 + }, + "additional_details": { + "max_retries": "1", + "timeout": "1200", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/b46321ef-c664-4379-878a-00c030df0f95_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "768712b8ea301863065680468962a43044d83acf5a644c570907e410644d1533", + "total_rows": 2 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/b46321ef-c664-4379-878a-00c030df0f95_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/b46321ef-c664-4379-878a-00c030df0f95_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..9e1f2f3e75f8f6dc8fe57d60f2f3160e720ea0e9 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/b46321ef-c664-4379-878a-00c030df0f95_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:768712b8ea301863065680468962a43044d83acf5a644c570907e410644d1533 +size 50943 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/ba52a7a3-e008-40d9-a892-97bb45b0e4f2.json b/data/hle/openai/gpt-5.2-2025-12-11/ba52a7a3-e008-40d9-a892-97bb45b0e4f2.json new file mode 100644 index 0000000000000000000000000000000000000000..9d290494a9922fa6fa36b3c587fd161c6f01250e --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/ba52a7a3-e008-40d9-a892-97bb45b0e4f2.json @@ -0,0 +1,112 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118302.881728", + "evaluation_timestamp": "1777387088.0", + "retrieved_timestamp": "1786118302.881728", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+3ep/cfc18407 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 3, + "sample_ids": [ + "hle_ebd62885", + "hle_a764505b", + "hle_34c69126" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1777387088.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5, + "uncertainty": { + "standard_error": { + "value": 0.5 + }, + "num_samples": 3 + }, + "details": { + "total_matched_trajectories": "3", + "stopping_reason_count_repetition_guard": "2", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "35693.67", + "total_total_tokens_target_model": "107081.0", + "avg_total_tokens_other_models": "3089.33", + "total_total_tokens_other_models": "9268.0", + "avg_total_tokens_all_models": "38783.00", + "total_total_tokens_all_models": "116349.0", + "avg_turn_count": "4.67", + "total_turn_count": "14.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/ba52a7a3-e008-40d9-a892-97bb45b0e4f2_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "c64de8a9f9a32f5c923c169037debb77bd3b2db96d64d2e082e75b1b7ba6629e", + "total_rows": 3 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/ba52a7a3-e008-40d9-a892-97bb45b0e4f2_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/ba52a7a3-e008-40d9-a892-97bb45b0e4f2_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..adfa80efd2e59a218da979e4f7d42b0c25798fb8 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/ba52a7a3-e008-40d9-a892-97bb45b0e4f2_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c64de8a9f9a32f5c923c169037debb77bd3b2db96d64d2e082e75b1b7ba6629e +size 127143 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/d33d96c0-d5dd-415e-8b5a-9172a0c677c1.json b/data/hle/openai/gpt-5.2-2025-12-11/d33d96c0-d5dd-415e-8b5a-9172a0c677c1.json new file mode 100644 index 0000000000000000000000000000000000000000..7b1e84e59fe6b9f2e11fa0eea7a7041f60ebd6a8 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/d33d96c0-d5dd-415e-8b5a-9172a0c677c1.json @@ -0,0 +1,114 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118307.214171", + "evaluation_timestamp": "1777348921.0", + "retrieved_timestamp": "1786118307.214171", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+3ep/e76f121e for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 5, + "sample_ids": [ + "hle_4418bea4", + "hle_ebd62885", + "hle_34c69126", + "hle_ee938b26", + "hle_c6adc1b4" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1777348921.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.3333333333333333, + "uncertainty": { + "standard_error": { + "value": 0.33333333333333337 + }, + "num_samples": 7 + }, + "details": { + "total_matched_trajectories": "7", + "stopping_reason_count_completed_on_successful_submit": "3", + "stopping_reason_count_repetition_guard": "4", + "avg_total_tokens_target_model": "41014.29", + "total_total_tokens_target_model": "287100.0", + "avg_total_tokens_other_models": "4941.71", + "total_total_tokens_other_models": "34592.0", + "avg_total_tokens_all_models": "45956.00", + "total_total_tokens_all_models": "321692.0", + "avg_turn_count": "5.14", + "total_turn_count": "36.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/d33d96c0-d5dd-415e-8b5a-9172a0c677c1_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "d82fde476390d434c6fd259f6ca3edd418c183c5c796bcb6055ec7106d46e4de", + "total_rows": 7 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/d33d96c0-d5dd-415e-8b5a-9172a0c677c1_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/d33d96c0-d5dd-415e-8b5a-9172a0c677c1_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..1414685435ae43acb497e46f3d189a5ba8325534 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/d33d96c0-d5dd-415e-8b5a-9172a0c677c1_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d82fde476390d434c6fd259f6ca3edd418c183c5c796bcb6055ec7106d46e4de +size 199911 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/da25e175-9be2-4dd7-b3f5-8d05e527c40c.json b/data/hle/openai/gpt-5.2-2025-12-11/da25e175-9be2-4dd7-b3f5-8d05e527c40c.json new file mode 100644 index 0000000000000000000000000000000000000000..26c8752a7dec49e3877939f0f4fee806d315b4cd --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/da25e175-9be2-4dd7-b3f5-8d05e527c40c.json @@ -0,0 +1,116 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118305.004463", + "evaluation_timestamp": "1778546664.0", + "retrieved_timestamp": "1786118305.004463", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+3ep/00e39d33 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 3, + "sample_ids": [ + "hle_28643173", + "hle_ebd62885", + "hle_d7b7c3dc" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1778546664.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5, + "uncertainty": { + "standard_error": { + "value": 0.2886751345948129 + }, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_with_submit": "1", + "stopping_reason_count_completed_on_successful_submit": "2", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "26041.00", + "total_total_tokens_target_model": "104164.0", + "avg_total_tokens_other_models": "2629.00", + "total_total_tokens_other_models": "10516.0", + "avg_total_tokens_all_models": "28670.00", + "total_total_tokens_all_models": "114680.0", + "avg_turn_count": "2.00", + "total_turn_count": "8.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/da25e175-9be2-4dd7-b3f5-8d05e527c40c_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "7d11d23d7b6bd96975f5a2a2b66897fb45559719f487e7f6d645829cc42bda9c", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/da25e175-9be2-4dd7-b3f5-8d05e527c40c_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/da25e175-9be2-4dd7-b3f5-8d05e527c40c_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..4356c225dfb9b681578d6c7b826d37e42a4347de --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/da25e175-9be2-4dd7-b3f5-8d05e527c40c_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d11d23d7b6bd96975f5a2a2b66897fb45559719f487e7f6d645829cc42bda9c +size 300174 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/dbdff093-b700-4125-aa46-99335898eb1c.json b/data/hle/openai/gpt-5.2-2025-12-11/dbdff093-b700-4125-aa46-99335898eb1c.json new file mode 100644 index 0000000000000000000000000000000000000000..df9153daedf0b2e823385a17b7dde8147b289ccb --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/dbdff093-b700-4125-aa46-99335898eb1c.json @@ -0,0 +1,110 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118263.889846", + "evaluation_timestamp": "1782222861.0", + "retrieved_timestamp": "1786118263.889846", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/57058abc for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 1, + "sample_ids": [ + "hle_c2ff3a43" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1782222861.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.0, + "uncertainty": { + "num_samples": 1 + }, + "details": { + "total_matched_trajectories": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "458333.00", + "total_total_tokens_target_model": "458333.0", + "avg_total_tokens_other_models": "4555.00", + "total_total_tokens_other_models": "4555.0", + "avg_total_tokens_all_models": "462888.00", + "total_total_tokens_all_models": "462888.0", + "avg_turn_count": "20.00", + "total_turn_count": "20.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "1200", + "max_connections": "16", + "max_tokens": "20000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/dbdff093-b700-4125-aa46-99335898eb1c_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "e23f9f6f92b07c6716a1b606f3feef9fbd464c91916b6de10f055b078d7ba3ac", + "total_rows": 1 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/dbdff093-b700-4125-aa46-99335898eb1c_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/dbdff093-b700-4125-aa46-99335898eb1c_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a68da619572f0e3e1997ea344da62da6cf714d2e --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/dbdff093-b700-4125-aa46-99335898eb1c_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e23f9f6f92b07c6716a1b606f3feef9fbd464c91916b6de10f055b078d7ba3ac +size 39653 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/e351f2b6-351c-42fc-a2f8-060c530b6c44.json b/data/hle/openai/gpt-5.2-2025-12-11/e351f2b6-351c-42fc-a2f8-060c530b6c44.json new file mode 100644 index 0000000000000000000000000000000000000000..86b04f80905772cfc57891d677fd961fbef3685f --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/e351f2b6-351c-42fc-a2f8-060c530b6c44.json @@ -0,0 +1,110 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118269.53361", + "evaluation_timestamp": "1776519736.0", + "retrieved_timestamp": "1786118269.53361", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/40afbcda for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 3, + "sample_ids": [ + "hle_baccd6da", + "hle_8770ab31", + "hle_6499b500" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1776519736.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.16666666666666666, + "uncertainty": { + "standard_error": { + "value": 0.16666666666666669 + }, + "num_samples": 6 + }, + "details": { + "total_matched_trajectories": "6", + "stopping_reason_count_repetition_guard": "5", + "stopping_reason_count_completed_on_successful_submit": "1", + "avg_total_tokens_target_model": "45430.83", + "total_total_tokens_target_model": "272585.0", + "avg_total_tokens_other_models": "5195.00", + "total_total_tokens_other_models": "31170.0", + "avg_total_tokens_all_models": "50625.83", + "total_total_tokens_all_models": "303755.0", + "avg_turn_count": "7.33", + "total_turn_count": "44.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "attempt_timeout": "1200", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "32000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/e351f2b6-351c-42fc-a2f8-060c530b6c44_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "890731ccdf10132f8f20660ea743a5406a669d482a5b142a2510ca41bf781c6a", + "total_rows": 6 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/e351f2b6-351c-42fc-a2f8-060c530b6c44_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/e351f2b6-351c-42fc-a2f8-060c530b6c44_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a0b77f55d483bffb262f6cb36fd7f3535b459858 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/e351f2b6-351c-42fc-a2f8-060c530b6c44_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:890731ccdf10132f8f20660ea743a5406a669d482a5b142a2510ca41bf781c6a +size 150832 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/f782ca99-3236-4c54-8d2c-98e0742c9cbb.json b/data/hle/openai/gpt-5.2-2025-12-11/f782ca99-3236-4c54-8d2c-98e0742c9cbb.json new file mode 100644 index 0000000000000000000000000000000000000000..a0df3a48d1b23f6f16777ddadcb4049cc4882001 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/f782ca99-3236-4c54-8d2c-98e0742c9cbb.json @@ -0,0 +1,104 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118258.63274", + "evaluation_timestamp": "1782322677.0", + "retrieved_timestamp": "1786118258.63274", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/0cf791b7 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 4, + "sample_ids": [ + "hle_376b8247", + "hle_70e5aebc", + "hle_34c69126", + "hle_c2ff3a43" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1782322677.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.75, + "uncertainty": { + "standard_error": { + "value": 0.25 + }, + "num_samples": 4 + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "1200", + "max_connections": "16", + "max_tokens": "20000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/f782ca99-3236-4c54-8d2c-98e0742c9cbb_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "9c06f9b9a9702cfb4905d89fa353c255158a9a53629b46c17ec70b427b31f7f2", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/f782ca99-3236-4c54-8d2c-98e0742c9cbb_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/f782ca99-3236-4c54-8d2c-98e0742c9cbb_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b66c00267eb6a6c1e8b7a232d621a992adc2f12f --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/f782ca99-3236-4c54-8d2c-98e0742c9cbb_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c06f9b9a9702cfb4905d89fa353c255158a9a53629b46c17ec70b427b31f7f2 +size 50622 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/faeb4130-496d-490c-a56c-aaf48e98b779.json b/data/hle/openai/gpt-5.2-2025-12-11/faeb4130-496d-490c-a56c-aaf48e98b779.json new file mode 100644 index 0000000000000000000000000000000000000000..032725ad0abde7e4e6c5932bd7a7e65400e6beb8 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/faeb4130-496d-490c-a56c-aaf48e98b779.json @@ -0,0 +1,115 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118267.547349", + "evaluation_timestamp": "1783094341.0", + "retrieved_timestamp": "1786118267.547349", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+2ep/2583114f for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 2, + "sample_ids": [ + "hle_44007b97", + "hle_b26bf8aa" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1783094341.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5, + "uncertainty": { + "standard_error": { + "value": 0.5 + }, + "num_samples": 4 + }, + "details": { + "total_matched_trajectories": "4", + "stopping_reason_count_completed_on_successful_submit": "2", + "stopping_reason_count_repetition_guard": "2", + "avg_total_tokens_target_model": "139799.75", + "total_total_tokens_target_model": "559199.0", + "avg_total_tokens_other_models": "3561.50", + "total_total_tokens_other_models": "14246.0", + "avg_total_tokens_all_models": "143361.25", + "total_total_tokens_all_models": "573445.0", + "avg_turn_count": "4.50", + "total_turn_count": "18.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "1", + "timeout": "1200", + "attempt_timeout": "3600", + "max_connections": "16", + "max_tokens": "64000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 1 + }, + "additional_details": { + "max_retries": "1", + "timeout": "1200", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/faeb4130-496d-490c-a56c-aaf48e98b779_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "67e4c7f7f7c8507bdac4b84c96e7fe696afab89dfb1a53a6fccdee84e2970cd1", + "total_rows": 4 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/faeb4130-496d-490c-a56c-aaf48e98b779_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/faeb4130-496d-490c-a56c-aaf48e98b779_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..07543c274acbe05e389b812fa133949b43b8054f --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/faeb4130-496d-490c-a56c-aaf48e98b779_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67e4c7f7f7c8507bdac4b84c96e7fe696afab89dfb1a53a6fccdee84e2970cd1 +size 71321 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/fcb5d5e1-4390-4be4-904e-b4ea752e6ad3.json b/data/hle/openai/gpt-5.2-2025-12-11/fcb5d5e1-4390-4be4-904e-b4ea752e6ad3.json new file mode 100644 index 0000000000000000000000000000000000000000..3f5a80e741da6c0dccd0b84f9e98c850540296a3 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/fcb5d5e1-4390-4be4-904e-b4ea752e6ad3.json @@ -0,0 +1,116 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118228.876657", + "evaluation_timestamp": "1778667069.0", + "retrieved_timestamp": "1786118228.876657", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/8fad8fb6 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 3, + "sample_ids": [ + "hle_9019c023", + "hle_d7b7c3dc", + "hle_a764505b" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1778667069.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.5, + "uncertainty": { + "standard_error": { + "value": 0.5 + }, + "num_samples": 2 + }, + "details": { + "total_matched_trajectories": "2", + "stopping_reason_count_completed_on_successful_submit": "1", + "stopping_reason_count_repetition_guard": "1", + "avg_total_tokens_target_model": "12018.50", + "total_total_tokens_target_model": "24037.0", + "avg_total_tokens_other_models": "2333.00", + "total_total_tokens_other_models": "4666.0", + "avg_total_tokens_all_models": "14351.50", + "total_total_tokens_all_models": "28703.0", + "avg_turn_count": "3.00", + "total_turn_count": "6.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "max_retries": "5", + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "16", + "max_tokens": "20000", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {}, + "max_attempts": 5 + }, + "additional_details": { + "max_retries": "5", + "timeout": "5400", + "max_connections": "16", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/fcb5d5e1-4390-4be4-904e-b4ea752e6ad3_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "2ab13ed9a164d9f91939e730966baf03e06c7482e01bb1fabd3eb9995c0a7aef", + "total_rows": 2 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/fcb5d5e1-4390-4be4-904e-b4ea752e6ad3_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/fcb5d5e1-4390-4be4-904e-b4ea752e6ad3_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..e2f79db5fb16fd7edd37c97fcbf85bdf62b3f763 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/fcb5d5e1-4390-4be4-904e-b4ea752e6ad3_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ab13ed9a164d9f91939e730966baf03e06c7482e01bb1fabd3eb9995c0a7aef +size 65442 diff --git a/data/hle/openai/gpt-5.2-2025-12-11/fe09a8b5-da67-4c61-a378-77092e96ae08.json b/data/hle/openai/gpt-5.2-2025-12-11/fe09a8b5-da67-4c61-a378-77092e96ae08.json new file mode 100644 index 0000000000000000000000000000000000000000..4f4ba0f9c05ab6a35cbb4a10e7c6bb8401412413 --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/fe09a8b5-da67-4c61-a378-77092e96ae08.json @@ -0,0 +1,141 @@ +{ + "schema_version": "0.3.0", + "evaluation_id": "hle/openai_gpt-5.2-2025-12-11/1786118231.431963", + "evaluation_timestamp": "1777292858.0", + "retrieved_timestamp": "1786118231.431963", + "source_metadata": { + "source_name": "inspect_ai", + "source_type": "evaluation_run", + "source_organization_name": "unknown", + "evaluator_relationship": "third_party" + }, + "eval_library": { + "name": "inspect_ai", + "version": "inspect_ai:0.3.206.dev19+g6beb95d9,inference_scaling_paper:0.1.0" + }, + "model_info": { + "name": "openai/gpt-5.2-2025-12-11", + "id": "openai/gpt-5.2-2025-12-11", + "developer": "openai", + "inference_platform": "openai", + "additional_details": { + "deployment_type": "externally_managed", + "model_availability": "closed_weights" + } + }, + "evaluation_results": [ + { + "evaluation_name": "mean on hle/S-adaptive+C/+1ep/160937f5 for scorer _llm_grader", + "source_data": { + "dataset_name": "hle", + "source_type": "hf_dataset", + "hf_repo": "cais/hle", + "samples_number": 32, + "sample_ids": [ + "hle_16388cc8", + "hle_9019c023", + "hle_c3ca0bda", + "hle_0291c889", + "hle_dab7fbb1", + "hle_5e235b48", + "hle_0ff76ff3", + "hle_5759e7d7", + "hle_1dc141bb", + "hle_e62ba103", + "hle_5f04b314", + "hle_d5a55900", + "hle_82f5c665", + "hle_cc7f25c0", + "hle_226e134b", + "hle_c1158fb8", + "hle_d0ea1019", + "hle_fcd42d0c", + "hle_b33012cf", + "hle_3697d2bc", + "hle_5e349785", + "hle_8b0384c4", + "hle_d190d8c2", + "hle_5671e386", + "hle_a634e89e", + "hle_670287ad", + "hle_298b5616", + "hle_e14d8c23", + "hle_510b6040", + "hle_aa343d51", + "hle_d52d3dae", + "hle_c0ed25f7" + ], + "additional_details": { + "shuffled": "False" + } + }, + "evaluation_timestamp": "1777292858.0", + "metric_config": { + "evaluation_description": "mean", + "lower_is_better": false, + "score_type": "continuous", + "min_score": 0.0, + "max_score": 1.0 + }, + "score_details": { + "score": 0.41379310344827586, + "uncertainty": { + "standard_error": { + "value": 0.0930760769837004 + }, + "num_samples": 29 + }, + "details": { + "total_matched_trajectories": "29", + "stopping_reason_count_repetition_guard": "17", + "stopping_reason_count_completed_on_successful_submit": "12", + "avg_total_tokens_target_model": "23765.45", + "total_total_tokens_target_model": "689198.0", + "avg_total_tokens_other_models": "3910.38", + "total_total_tokens_other_models": "113401.0", + "avg_total_tokens_all_models": "27675.83", + "total_total_tokens_all_models": "802599.0", + "avg_turn_count": "7.83", + "total_turn_count": "227.0" + } + }, + "generation_config": { + "generation_args": { + "reasoning": true, + "agentic_eval_config": { + "available_tools": [] + }, + "eval_plan": { + "name": "plan", + "steps": [ + "{\"solver\": \"react\", \"params\": {\"name\": null, \"description\": null, \"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"model\": null, \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"retry_refusals\": null, \"compaction\": \"CompactionSummary\", \"truncation\": \"disabled\", \"approval\": null}, \"params_passed\": {\"prompt\": [\"\", \"\\nYou are part of a multi-agent system designed to make agent coordination and execution easy. Agents uses two primary abstraction: **Agents** and **Handoffs**. An agent encompasses instructions and tools and can hand off a conversation to another agent when appropriate. Handoffs are achieved by calling a handoff function,generally named `transfer_to_`. Transfers between agents are handled seamlessly in the background; do not mention or draw attention to these transfers in your conversation with the user.\\n\", \"\\nYou are a helpful assistant attempting to submit the best possible answer. You have several tools available to help with finding the answer. You will see the result of tool calls right after sending the message. If you need to perform multiple actions, you can always send more messages with additional tool calls. Do some reasoning before your actions, describing what tool calls you are going to use and how they fit into your plan.\\n\", \"\\nWhen you have completed the task and have an answer, call the {submit}() tool to report it.\\n\"], \"tools\": [], \"attempts\": [999, \"wrapper\", \"to_float\"], \"submit\": null, \"on_continue\": \"_guard\", \"compaction\": \"CompactionSummary\"}}" + ], + "config": { + "timeout": "5400", + "attempt_timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"", + "reasoning_tokens": "16000" + } + }, + "eval_limits": { + "token_limit": 5000000 + }, + "sandbox": {} + }, + "additional_details": { + "timeout": "5400", + "max_connections": "10", + "reasoning_effort": "\"high\"" + } + } + } + ], + "detailed_evaluation_results": { + "format": "jsonl", + "file_path": "data/hle/openai/gpt-5.2-2025-12-11/fe09a8b5-da67-4c61-a378-77092e96ae08_samples.jsonl", + "hash_algorithm": "sha256", + "checksum": "1ebca04b8f7a21bf793f1ebfd404d4e689858a47010a93d9a1fecc3e286eeec8", + "total_rows": 29 + } +} \ No newline at end of file diff --git a/data/hle/openai/gpt-5.2-2025-12-11/fe09a8b5-da67-4c61-a378-77092e96ae08_samples.jsonl b/data/hle/openai/gpt-5.2-2025-12-11/fe09a8b5-da67-4c61-a378-77092e96ae08_samples.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..23aaf59639fad41034330c77823a6fe1fd5d83cc --- /dev/null +++ b/data/hle/openai/gpt-5.2-2025-12-11/fe09a8b5-da67-4c61-a378-77092e96ae08_samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ebca04b8f7a21bf793f1ebfd404d4e689858a47010a93d9a1fecc3e286eeec8 +size 794389