{"api_version":"v1","generated_at":"2026-10-07T17:52:57Z","data_retrieved_at":"2026-10-07","harnesses":["artificial-analysis"],"counts":{"models":85,"benchmarks":24,"measured":396,"estimated":207,"mappings":34,"confidence":{"high":58,"medium":82,"low":67}},"scores":[{"model":"claude-fable-5-1-max-with-fallback","benchmark":"terminal-bench/2.1","score":0.914,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"terminal-bench/2.1","score":0.91,"source":"measured","estimate":null},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"terminal-bench/2.1","score":0.899,"source":"measured","estimate":null},{"model":"gpt-6-astra-high","benchmark":"terminal-bench/2.1","score":0.899,"source":"measured","estimate":null},{"model":"gpt-6-astra-medium","benchmark":"terminal-bench/2.1","score":0.895,"source":"measured","estimate":null},{"model":"gpt-5-6-sol-xhigh","benchmark":"terminal-bench/2.1","score":0.895,"source":"measured","estimate":null},{"model":"qwen3-8-max-0902","benchmark":"terminal-bench/2.1","score":0.888,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"terminal-bench/2.1","score":0.884,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"terminal-bench/2.1","score":0.876,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"terminal-bench/2.1","score":0.85,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"terminal-bench/2.1","score":0.843,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"terminal-bench/2.1","score":0.843,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"terminal-bench/2.1","score":0.839,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"terminal-bench/2.1","score":0.798,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"terminal-bench/2.1","score":0.719,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"terminal-bench/2.1","score":0.652,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"terminal-bench/2.1","score":0.551,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"terminal-bench/2.1","score":0.539,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"terminal-bench/2.1","score":0.517,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"terminal-bench/2.1","score":0.883091,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.96,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.636}]}},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"terminal-bench/2.1","score":0.882377,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.96,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.596}]}},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"terminal-bench/2.1","score":0.882377,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.96,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.596}]}},{"model":"gpt-6-astra-xhigh","benchmark":"terminal-bench/2.1","score":0.882377,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.96,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.596}]}},{"model":"gemini-4-argon-high","benchmark":"terminal-bench/2.1","score":0.881881,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.571}]}},{"model":"claude-sonnet-5-5-xhigh-with-fallback","benchmark":"terminal-bench/2.1","score":0.881881,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.571}]}},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"terminal-bench/2.1","score":0.881776,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.566}]}},{"model":"gpt-6-1-sol-max","benchmark":"terminal-bench/2.1","score":0.88167,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.561}]}},{"model":"gpt-6-sol-max","benchmark":"terminal-bench/2.1","score":0.878355,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.439}]}},{"model":"mimo-v2-6-pro","benchmark":"terminal-bench/2.1","score":0.874427,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.348}]}},{"model":"step-5-preview","benchmark":"terminal-bench/2.1","score":0.873582,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.333}]}},{"model":"deepseek-v4-1-flash-max","benchmark":"terminal-bench/2.1","score":0.868881,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.268}]}},{"model":"mistral-large-4-preview","benchmark":"terminal-bench/2.1","score":0.868881,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.268}]}},{"model":"grok-4-7-xhigh","benchmark":"terminal-bench/2.1","score":0.867958,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.258}]}},{"model":"gpt-6-luna-max","benchmark":"terminal-bench/2.1","score":0.84343,"source":"estimated","estimate":{"confidence":"high","error_pp":3.96,"reasons":[],"extrapolated":false,"method":"mm_offset","method_name":"Michaelis–Menten + offset","kind":"univariate","mapping_id":31,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.126}]}},{"model":"claude-opus-5-5-medium-with-fallback","benchmark":"terminal-bench/2.1","score":0.894286,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.03,"reasons":["cross-validated error ±7.0 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":33,"inputs":[{"benchmark":"terminal-bench-science/0.1","score":0.433}]}},{"model":"claude-sonnet-5-5-high-with-fallback","benchmark":"terminal-bench/2.1","score":0.890572,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.03,"reasons":["cross-validated error ±7.0 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":33,"inputs":[{"benchmark":"terminal-bench-science/0.1","score":0.343}]}},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"terminal-bench/4.0","score":0.52,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"terminal-bench/4.0","score":0.551,"source":"measured","estimate":null},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"terminal-bench/4.0","score":0.478073,"source":"estimated","estimate":{"confidence":"low","error_pp":11.55,"reasons":["large cross-validated error ±11.6 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":30,"inputs":[{"benchmark":"terminal-bench/2.1","score":0.899}]}},{"model":"gpt-6-astra-high","benchmark":"terminal-bench/4.0","score":0.478073,"source":"estimated","estimate":{"confidence":"low","error_pp":11.55,"reasons":["large cross-validated error ±11.6 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":30,"inputs":[{"benchmark":"terminal-bench/2.1","score":0.899}]}},{"model":"gpt-6-astra-medium","benchmark":"terminal-bench/4.0","score":0.462831,"source":"estimated","estimate":{"confidence":"low","error_pp":11.55,"reasons":["large cross-validated error ±11.6 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":30,"inputs":[{"benchmark":"terminal-bench/2.1","score":0.895}]}},{"model":"gpt-5-6-sol-xhigh","benchmark":"terminal-bench/4.0","score":0.462831,"source":"estimated","estimate":{"confidence":"low","error_pp":11.55,"reasons":["large cross-validated error ±11.6 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":30,"inputs":[{"benchmark":"terminal-bench/2.1","score":0.895}]}},{"model":"qwen3-8-max-0902","benchmark":"terminal-bench/4.0","score":0.389,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"terminal-bench/4.0","score":0.591,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"terminal-bench/4.0","score":0.197,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"terminal-bench/4.0","score":0.126,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"terminal-bench/4.0","score":0.333,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"terminal-bench/4.0","score":0.328,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"terminal-bench/4.0","score":0.419,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"terminal-bench/4.0","score":0.056,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"terminal-bench/4.0","score":0.015,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"terminal-bench/4.0","score":0.02,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"terminal-bench/4.0","score":0.01,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"terminal-bench/4.0","score":0.005,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"terminal-bench/4.0","score":0.005,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"terminal-bench/4.0","score":0.636,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"terminal-bench/4.0","score":0.596,"source":"measured","estimate":null},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"terminal-bench/4.0","score":0.596,"source":"measured","estimate":null},{"model":"gpt-6-astra-xhigh","benchmark":"terminal-bench/4.0","score":0.596,"source":"measured","estimate":null},{"model":"gemini-4-argon-high","benchmark":"terminal-bench/4.0","score":0.571,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-xhigh-with-fallback","benchmark":"terminal-bench/4.0","score":0.571,"source":"measured","estimate":null},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"terminal-bench/4.0","score":0.566,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-max","benchmark":"terminal-bench/4.0","score":0.561,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"terminal-bench/4.0","score":0.439,"source":"measured","estimate":null},{"model":"mimo-v2-6-pro","benchmark":"terminal-bench/4.0","score":0.348,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"terminal-bench/4.0","score":0.333,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"terminal-bench/4.0","score":0.268,"source":"measured","estimate":null},{"model":"mistral-large-4-preview","benchmark":"terminal-bench/4.0","score":0.268,"source":"measured","estimate":null},{"model":"grok-4-7-xhigh","benchmark":"terminal-bench/4.0","score":0.258,"source":"measured","estimate":null},{"model":"gpt-6-luna-max","benchmark":"terminal-bench/4.0","score":0.126,"source":"measured","estimate":null},{"model":"claude-opus-5-5-medium-with-fallback","benchmark":"terminal-bench/4.0","score":0.535507,"source":"estimated","estimate":{"confidence":"medium","error_pp":8.23,"reasons":["cross-validated error ±8.2 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":34,"inputs":[{"benchmark":"terminal-bench-science/0.1","score":0.433}]}},{"model":"claude-sonnet-5-5-high-with-fallback","benchmark":"terminal-bench/4.0","score":0.500107,"source":"estimated","estimate":{"confidence":"medium","error_pp":8.23,"reasons":["cross-validated error ±8.2 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":34,"inputs":[{"benchmark":"terminal-bench-science/0.1","score":0.343}]}},{"model":"minimax-m3","benchmark":"terminal-bench-hard/1.0","score":0.424,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"terminal-bench-hard/1.0","score":0.364,"source":"measured","estimate":null},{"model":"gpt-5-6-sol-max","benchmark":"terminal-bench-hard/1.0","score":0.659,"source":"measured","estimate":null},{"model":"claude-fable-5-with-fallback","benchmark":"terminal-bench-hard/1.0","score":0.629,"source":"measured","estimate":null},{"model":"gpt-5-6-sol-medium","benchmark":"terminal-bench-hard/1.0","score":0.629,"source":"measured","estimate":null},{"model":"gpt-5-6-terra-xhigh","benchmark":"terminal-bench-hard/1.0","score":0.629,"source":"measured","estimate":null},{"model":"gpt-5-6-sol-high","benchmark":"terminal-bench-hard/1.0","score":0.621,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.433,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.493538,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.551}]}},{"model":"qwen3-8-max-0902","benchmark":"terminal-bench-science/0.1","score":0.119,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"terminal-bench-science/0.1","score":0.633,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"terminal-bench-science/0.1","score":0.1,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"terminal-bench-science/0.1","score":0.026191,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.126}]}},{"model":"muse-spark-1-3-max","benchmark":"terminal-bench-science/0.1","score":0.11,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"terminal-bench-science/0.1","score":0.048,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"terminal-bench-science/0.1","score":0.095,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"terminal-bench-science/0.1","score":0.005,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"terminal-bench-science/0.1","score":0.026028,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.015}]}},{"model":"minimax-m3","benchmark":"terminal-bench-science/0.1","score":0.005,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"terminal-bench-science/0.1","score":0,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"terminal-bench-science/0.1","score":0.026028,"source":"estimated","estimate":{"confidence":"low","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.005}]}},{"model":"muse-glimmer-high","benchmark":"terminal-bench-science/0.1","score":0.026028,"source":"estimated","estimate":{"confidence":"low","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.005}]}},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.533,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.59,"source":"measured","estimate":null},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.619,"source":"measured","estimate":null},{"model":"gpt-6-astra-xhigh","benchmark":"terminal-bench-science/0.1","score":0.578733,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.596}]}},{"model":"gemini-4-argon-high","benchmark":"terminal-bench-science/0.1","score":0.533655,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.571}]}},{"model":"claude-sonnet-5-5-xhigh-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.524,"source":"measured","estimate":null},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.49,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-max","benchmark":"terminal-bench-science/0.1","score":0.581,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"terminal-bench-science/0.1","score":0.3,"source":"measured","estimate":null},{"model":"mimo-v2-6-pro","benchmark":"terminal-bench-science/0.1","score":0.057,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"terminal-bench-science/0.1","score":0.024,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"terminal-bench-science/0.1","score":0.09,"source":"measured","estimate":null},{"model":"mistral-large-4-preview","benchmark":"terminal-bench-science/0.1","score":0.040832,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.268}]}},{"model":"grok-4-7-xhigh","benchmark":"terminal-bench-science/0.1","score":0.037857,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.51,"reasons":["cross-validated error ±6.5 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":32,"inputs":[{"benchmark":"terminal-bench/4.0","score":0.258}]}},{"model":"gpt-6-luna-max","benchmark":"terminal-bench-science/0.1","score":0.086,"source":"measured","estimate":null},{"model":"claude-opus-5-5-medium-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.433,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-high-with-fallback","benchmark":"terminal-bench-science/0.1","score":0.343,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"gpqa/diamond","score":0.937,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"gpqa/diamond","score":0.955954,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.587}]}},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"gpqa/diamond","score":0.951539,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.559}]}},{"model":"gpt-6-astra-high","benchmark":"gpqa/diamond","score":0.949,"source":"measured","estimate":null},{"model":"qwen3-8-max-0902","benchmark":"gpqa/diamond","score":0.928,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"gpqa/diamond","score":0.961,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"gpqa/diamond","score":0.953,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"gpqa/diamond","score":0.935,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"gpqa/diamond","score":0.935,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"gpqa/diamond","score":0.912,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"gpqa/diamond","score":0.917,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"gpqa/diamond","score":0.905,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"gpqa/diamond","score":0.873,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"gpqa/diamond","score":0.929,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"gpqa/diamond","score":0.872,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"gpqa/diamond","score":0.867,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"gpqa/diamond","score":0.835,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"gpqa/diamond","score":0.950033,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.55}]}},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"gpqa/diamond","score":0.959865,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.31,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.614}]}},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"gpqa/diamond","score":0.95411,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.575}]}},{"model":"gpt-6-astra-xhigh","benchmark":"gpqa/diamond","score":0.963,"source":"measured","estimate":null},{"model":"gemini-4-argon-high","benchmark":"gpqa/diamond","score":0.953479,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.571}]}},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"gpqa/diamond","score":0.951042,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.556}]}},{"model":"gpt-6-1-sol-max","benchmark":"gpqa/diamond","score":0.946342,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.529}]}},{"model":"gpt-6-sol-max","benchmark":"gpqa/diamond","score":0.936394,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.479}]}},{"model":"mimo-v2-6-pro","benchmark":"gpqa/diamond","score":0.939567,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.494}]}},{"model":"step-5-preview","benchmark":"gpqa/diamond","score":0.933269,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.465}]}},{"model":"deepseek-v4-1-flash-max","benchmark":"gpqa/diamond","score":0.913839,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.392}]}},{"model":"mistral-large-4-preview","benchmark":"gpqa/diamond","score":0.899523,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.35}]}},{"model":"grok-4-7-xhigh","benchmark":"gpqa/diamond","score":0.924938,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.431}]}},{"model":"gpt-6-luna-max","benchmark":"gpqa/diamond","score":0.91164,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.385}]}},{"model":"gpt-5-6-sol-max","benchmark":"gpqa/diamond","score":0.941,"source":"measured","estimate":null},{"model":"claude-fable-5-with-fallback","benchmark":"gpqa/diamond","score":0.950875,"source":"estimated","estimate":{"confidence":"high","error_pp":1.31,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":22,"inputs":[{"benchmark":"hle/1.0","score":0.555}]}},{"model":"grok-4-6-high","benchmark":"gpqa/diamond","score":0.949,"source":"measured","estimate":null},{"model":"gemini-3-7-flash-high","benchmark":"gpqa/diamond","score":0.945,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-xhigh","benchmark":"gpqa/diamond","score":0.941631,"source":"estimated","estimate":{"confidence":"high","error_pp":2.26,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":9,"inputs":[{"benchmark":"critpt/1.0","score":0.317}]}},{"model":"gemini-3-pro-preview-high","benchmark":"mmlu-pro/1.0","score":0.898,"source":"measured","estimate":null},{"model":"gemini-3-pro-preview-low","benchmark":"mmlu-pro/1.0","score":0.895,"source":"measured","estimate":null},{"model":"claude-opus-4-5","benchmark":"mmlu-pro/1.0","score":0.895,"source":"measured","estimate":null},{"model":"gemini-3-flash","benchmark":"mmlu-pro/1.0","score":0.89,"source":"measured","estimate":null},{"model":"claude-opus-4-5-non-reasoning","benchmark":"mmlu-pro/1.0","score":0.889,"source":"measured","estimate":null},{"model":"gemini-3-pro-preview-high","benchmark":"global-mmlu-lite/1.0","score":0.922,"source":"measured","estimate":null},{"model":"gemini-3-flash","benchmark":"global-mmlu-lite/1.0","score":0.914,"source":"measured","estimate":null},{"model":"gemini-3-1-pro-preview","benchmark":"global-mmlu-lite/1.0","score":0.932,"source":"measured","estimate":null},{"model":"gemini-3-flash-non-reasoning","benchmark":"global-mmlu-lite/1.0","score":0.927,"source":"measured","estimate":null},{"model":"claude-opus-4-6-max","benchmark":"global-mmlu-lite/1.0","score":0.922,"source":"measured","estimate":null},{"model":"gpt-5-high","benchmark":"math-500/1.0","score":0.994,"source":"measured","estimate":null},{"model":"o3","benchmark":"math-500/1.0","score":0.992,"source":"measured","estimate":null},{"model":"grok-3-mini-reasoning-high","benchmark":"math-500/1.0","score":0.992,"source":"measured","estimate":null},{"model":"gpt-5-medium","benchmark":"math-500/1.0","score":0.991,"source":"measured","estimate":null},{"model":"claude-4-sonnet","benchmark":"math-500/1.0","score":0.991,"source":"measured","estimate":null},{"model":"gemini-3-flash","benchmark":"aime-2025/2025","score":0.97,"source":"measured","estimate":null},{"model":"gpt-5-2-xhigh","benchmark":"aime-2025/2025","score":0.99,"source":"measured","estimate":null},{"model":"gpt-5-codex-high","benchmark":"aime-2025/2025","score":0.987,"source":"measured","estimate":null},{"model":"gpt-5-2-medium","benchmark":"aime-2025/2025","score":0.967,"source":"measured","estimate":null},{"model":"deepseek-v3-2-speciale","benchmark":"aime-2025/2025","score":0.967,"source":"measured","estimate":null},{"model":"gemini-3-pro-preview-high","benchmark":"livecodebench/1.0","score":0.917,"source":"measured","estimate":null},{"model":"gemini-3-flash","benchmark":"livecodebench/1.0","score":0.908,"source":"measured","estimate":null},{"model":"gpt-5-2-medium","benchmark":"livecodebench/1.0","score":0.894,"source":"measured","estimate":null},{"model":"deepseek-v3-2-speciale","benchmark":"livecodebench/1.0","score":0.896,"source":"measured","estimate":null},{"model":"glm-4-7","benchmark":"livecodebench/1.0","score":0.894,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"scicode/1.0","score":0.631,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"scicode/1.0","score":0.609,"source":"measured","estimate":null},{"model":"qwen3-8-max-0902","benchmark":"scicode/1.0","score":0.521,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"scicode/1.0","score":0.565,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"scicode/1.0","score":0.566,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"scicode/1.0","score":0.595,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"scicode/1.0","score":0.588,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"scicode/1.0","score":0.516,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"scicode/1.0","score":0.59,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"scicode/1.0","score":0.466,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"scicode/1.0","score":0.429,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"scicode/1.0","score":0.471,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"scicode/1.0","score":0.47,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"scicode/1.0","score":0.403,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"scicode/1.0","score":0.449,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"scicode/1.0","score":0.61,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"scicode/1.0","score":0.669,"source":"measured","estimate":null},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"scicode/1.0","score":0.65,"source":"measured","estimate":null},{"model":"gemini-4-argon-high","benchmark":"scicode/1.0","score":0.618,"source":"measured","estimate":null},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"scicode/1.0","score":0.604,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-max","benchmark":"scicode/1.0","score":0.542,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"scicode/1.0","score":0.576,"source":"measured","estimate":null},{"model":"mimo-v2-6-pro","benchmark":"scicode/1.0","score":0.609,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"scicode/1.0","score":0.589,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"scicode/1.0","score":0.519,"source":"measured","estimate":null},{"model":"mistral-large-4-preview","benchmark":"scicode/1.0","score":0.542,"source":"measured","estimate":null},{"model":"grok-4-7-xhigh","benchmark":"scicode/1.0","score":0.574,"source":"measured","estimate":null},{"model":"gpt-6-luna-max","benchmark":"scicode/1.0","score":0.546,"source":"measured","estimate":null},{"model":"claude-fable-5-with-fallback","benchmark":"scicode/1.0","score":0.61,"source":"measured","estimate":null},{"model":"gemini-3-7-flash-medium","benchmark":"scicode/1.0","score":0.598,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"critpt/1.0","score":0.297,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"critpt/1.0","score":0.311,"source":"measured","estimate":null},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"critpt/1.0","score":0.302091,"source":"estimated","estimate":{"confidence":"high","error_pp":3.48,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":21,"inputs":[{"benchmark":"hle/1.0","score":0.559}]}},{"model":"gpt-6-astra-high","benchmark":"critpt/1.0","score":0.259156,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.8,"reasons":["cross-validated error ±6.8 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":15,"inputs":[{"benchmark":"gpqa/diamond","score":0.949}]}},{"model":"qwen3-8-max-0902","benchmark":"critpt/1.0","score":0.177,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"critpt/1.0","score":0.317,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"critpt/1.0","score":0.183,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"critpt/1.0","score":0.234,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"critpt/1.0","score":0.249,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"critpt/1.0","score":0.154,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"critpt/1.0","score":0.191,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"critpt/1.0","score":0.054,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"critpt/1.0","score":0.046,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"critpt/1.0","score":0.037,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"critpt/1.0","score":0.054,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"critpt/1.0","score":0.031,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"critpt/1.0","score":0.026,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"critpt/1.0","score":0.314,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"critpt/1.0","score":0.317,"source":"measured","estimate":null},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"critpt/1.0","score":0.317,"source":"measured","estimate":null},{"model":"gpt-6-astra-xhigh","benchmark":"critpt/1.0","score":0.314,"source":"measured","estimate":null},{"model":"gemini-4-argon-high","benchmark":"critpt/1.0","score":0.271,"source":"measured","estimate":null},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"critpt/1.0","score":0.300444,"source":"estimated","estimate":{"confidence":"high","error_pp":3.48,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":21,"inputs":[{"benchmark":"hle/1.0","score":0.556}]}},{"model":"gpt-6-1-sol-max","benchmark":"critpt/1.0","score":0.317,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"critpt/1.0","score":0.309,"source":"measured","estimate":null},{"model":"mimo-v2-6-pro","benchmark":"critpt/1.0","score":0.266,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"critpt/1.0","score":0.209,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"critpt/1.0","score":0.143,"source":"measured","estimate":null},{"model":"mistral-large-4-preview","benchmark":"critpt/1.0","score":0.106,"source":"measured","estimate":null},{"model":"grok-4-7-xhigh","benchmark":"critpt/1.0","score":0.177,"source":"measured","estimate":null},{"model":"gpt-6-luna-max","benchmark":"critpt/1.0","score":0.194,"source":"measured","estimate":null},{"model":"gpt-5-6-sol-max","benchmark":"critpt/1.0","score":0.323,"source":"measured","estimate":null},{"model":"claude-fable-5-with-fallback","benchmark":"critpt/1.0","score":0.299886,"source":"estimated","estimate":{"confidence":"high","error_pp":3.48,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":21,"inputs":[{"benchmark":"hle/1.0","score":0.555}]}},{"model":"grok-4-6-high","benchmark":"critpt/1.0","score":0.259156,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.8,"reasons":["cross-validated error ±6.8 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":15,"inputs":[{"benchmark":"gpqa/diamond","score":0.949}]}},{"model":"gemini-3-7-flash-high","benchmark":"critpt/1.0","score":0.244582,"source":"estimated","estimate":{"confidence":"medium","error_pp":6.8,"reasons":["cross-validated error ±6.8 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":15,"inputs":[{"benchmark":"gpqa/diamond","score":0.945}]}},{"model":"gpt-6-1-sol-xhigh","benchmark":"critpt/1.0","score":0.317,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"gdp-pdf/1.0","score":0.262,"source":"measured","estimate":null},{"model":"gpt-6-astra-high","benchmark":"gdp-pdf/1.0","score":0.31,"source":"measured","estimate":null},{"model":"gpt-6-astra-medium","benchmark":"gdp-pdf/1.0","score":0.304,"source":"measured","estimate":null},{"model":"qwen3-8-max-0902","benchmark":"gdp-pdf/1.0","score":0.228,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"gdp-pdf/1.0","score":0.31,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"gdp-pdf/1.0","score":0.21,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"gdp-pdf/1.0","score":0.22,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"gdp-pdf/1.0","score":0.266,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"gdp-pdf/1.0","score":0.154,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"gdp-pdf/1.0","score":0.112,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"gdp-pdf/1.0","score":0.166,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"gdp-pdf/1.0","score":0.074,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"gdp-pdf/1.0","score":0.098,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"gdp-pdf/1.0","score":0.128,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"gdp-pdf/1.0","score":0.05,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"gdp-pdf/1.0","score":0.1,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"gdp-pdf/1.0","score":0.258,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"gdp-pdf/1.0","score":0.262,"source":"measured","estimate":null},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"gdp-pdf/1.0","score":0.291432,"source":"estimated","estimate":{"confidence":"high","error_pp":4.02,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":25,"inputs":[{"benchmark":"mmmu-pro/1.0","score":0.87}]}},{"model":"gpt-6-astra-xhigh","benchmark":"gdp-pdf/1.0","score":0.322,"source":"measured","estimate":null},{"model":"gemini-4-argon-high","benchmark":"gdp-pdf/1.0","score":0.218,"source":"measured","estimate":null},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"gdp-pdf/1.0","score":0.276401,"source":"estimated","estimate":{"confidence":"high","error_pp":4.02,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":25,"inputs":[{"benchmark":"mmmu-pro/1.0","score":0.86}]}},{"model":"gpt-6-1-sol-max","benchmark":"gdp-pdf/1.0","score":0.31,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"gdp-pdf/1.0","score":0.252,"source":"measured","estimate":null},{"model":"mimo-v2-6-pro","benchmark":"gdp-pdf/1.0","score":0.192,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"gdp-pdf/1.0","score":0.148,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"gdp-pdf/1.0","score":0.128,"source":"measured","estimate":null},{"model":"mistral-large-4-preview","benchmark":"gdp-pdf/1.0","score":0.186,"source":"measured","estimate":null},{"model":"grok-4-7-xhigh","benchmark":"gdp-pdf/1.0","score":0.2,"source":"measured","estimate":null},{"model":"gpt-6-luna-max","benchmark":"gdp-pdf/1.0","score":0.228,"source":"measured","estimate":null},{"model":"claude-opus-5-5-medium-with-fallback","benchmark":"gdp-pdf/1.0","score":0.276401,"source":"estimated","estimate":{"confidence":"high","error_pp":4.02,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":25,"inputs":[{"benchmark":"mmmu-pro/1.0","score":0.86}]}},{"model":"gpt-6-1-sol-xhigh","benchmark":"gdp-pdf/1.0","score":0.318,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-high","benchmark":"gdp-pdf/1.0","score":0.32,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"tau-bank/3-banking","score":0.472,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"tau-bank/3-banking","score":0.404552,"source":"estimated","estimate":{"confidence":"low","error_pp":10.41,"reasons":["large cross-validated error ±10.4 pp","fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":20,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.933}]}},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"tau-bank/3-banking","score":0.399386,"source":"estimated","estimate":{"confidence":"low","error_pp":10.41,"reasons":["large cross-validated error ±10.4 pp","fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":20,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.93}]}},{"model":"gpt-6-astra-high","benchmark":"tau-bank/3-banking","score":0.474,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.03,"reasons":["cross-validated error ±5.0 pp"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":8,"inputs":[{"benchmark":"automationbench/1.0","score":0.666}]}},{"model":"qwen3-8-max-0902","benchmark":"tau-bank/3-banking","score":0.478,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"tau-bank/3-banking","score":0.414,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"tau-bank/3-banking","score":0.449,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"tau-bank/3-banking","score":0.46,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"tau-bank/3-banking","score":0.505,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"tau-bank/3-banking","score":0.472,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"tau-bank/3-banking","score":0.503,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"tau-bank/3-banking","score":0.48,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"tau-bank/3-banking","score":0.342,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"tau-bank/3-banking","score":0.153,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"tau-bank/3-banking","score":0.291,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"tau-bank/3-banking","score":0.142,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"tau-bank/3-banking","score":0.235,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"tau-bank/3-banking","score":0.482876,"source":"estimated","estimate":{"confidence":"medium","error_pp":4.54,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":4,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.575}]}},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"tau-bank/3-banking","score":0.490421,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.62,"reasons":["mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.382}]}},{"model":"gpt-6-astra-xhigh","benchmark":"tau-bank/3-banking","score":0.474,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.03,"reasons":["cross-validated error ±5.0 pp"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":8,"inputs":[{"benchmark":"automationbench/1.0","score":0.672}]}},{"model":"gemini-4-argon-high","benchmark":"tau-bank/3-banking","score":0.474,"source":"estimated","estimate":{"confidence":"low","error_pp":5.03,"reasons":["cross-validated error ±5.0 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":8,"inputs":[{"benchmark":"automationbench/1.0","score":0.775}]}},{"model":"gpt-6-1-sol-max","benchmark":"tau-bank/3-banking","score":0.457753,"source":"estimated","estimate":{"confidence":"medium","error_pp":4.54,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":4,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.5}]}},{"model":"gpt-6-sol-max","benchmark":"tau-bank/3-banking","score":0.460372,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.62,"reasons":["mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.494}]}},{"model":"mimo-v2-6-pro","benchmark":"tau-bank/3-banking","score":0.474,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.03,"reasons":["cross-validated error ±5.0 pp"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":8,"inputs":[{"benchmark":"automationbench/1.0","score":0.586}]}},{"model":"step-5-preview","benchmark":"tau-bank/3-banking","score":0.443737,"source":"estimated","estimate":{"confidence":"low","error_pp":2.62,"reasons":["extrapolated beyond the scores the mapping was fitted on","mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.556}]}},{"model":"deepseek-v4-1-flash-max","benchmark":"tau-bank/3-banking","score":0.467079,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.62,"reasons":["mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.469}]}},{"model":"mistral-large-4-preview","benchmark":"tau-bank/3-banking","score":0.474,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.03,"reasons":["cross-validated error ±5.0 pp"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":8,"inputs":[{"benchmark":"automationbench/1.0","score":0.599}]}},{"model":"grok-4-7-xhigh","benchmark":"tau-bank/3-banking","score":0.479958,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.62,"reasons":["mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.421}]}},{"model":"gpt-6-luna-max","benchmark":"tau-bank/3-banking","score":0.474,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.03,"reasons":["cross-validated error ±5.0 pp"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":8,"inputs":[{"benchmark":"automationbench/1.0","score":0.532}]}},{"model":"gpt-5-6-sol-max","benchmark":"tau-bank/3-banking","score":0.442127,"source":"estimated","estimate":{"confidence":"low","error_pp":2.62,"reasons":["extrapolated beyond the scores the mapping was fitted on","mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.562}]}},{"model":"claude-fable-5-with-fallback","benchmark":"tau-bank/3-banking","score":0.453307,"source":"estimated","estimate":{"confidence":"medium","error_pp":4.54,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":4,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.488}]}},{"model":"grok-4-6-high","benchmark":"tau-bank/3-banking","score":0.507,"source":"measured","estimate":null},{"model":"gemini-3-7-flash-high","benchmark":"tau-bank/3-banking","score":0.490351,"source":"estimated","estimate":{"confidence":"low","error_pp":4.54,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":4,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.6}]}},{"model":"gemini-3-7-flash-medium","benchmark":"tau-bank/3-banking","score":0.503986,"source":"estimated","estimate":{"confidence":"low","error_pp":11.4,"reasons":["large cross-validated error ±11.4 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":13,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.504}]}},{"model":"gpt-6-1-sol-xhigh","benchmark":"tau-bank/3-banking","score":0.474,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.03,"reasons":["cross-validated error ±5.0 pp"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":8,"inputs":[{"benchmark":"automationbench/1.0","score":0.666}]}},{"model":"qwen3-8-max","benchmark":"tau-bank/3-banking","score":0.513,"source":"measured","estimate":null},{"model":"qwen3-8-2-4t-a95b","benchmark":"tau-bank/3-banking","score":0.491,"source":"measured","estimate":null},{"model":"qwen3-8-27b-medium","benchmark":"tau-bank/3-banking","score":0.474,"source":"measured","estimate":null},{"model":"muse-spark-1-3-xhigh","benchmark":"tau-bank/3-banking","score":0.472,"source":"measured","estimate":null},{"model":"claude-opus-5-max","benchmark":"tau-bank/3-banking","score":0.471023,"source":"estimated","estimate":{"confidence":"medium","error_pp":4.54,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":4,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.538}]}},{"model":"gpt-5-5-xhigh","benchmark":"tau-bank/3-banking","score":0.470031,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.62,"reasons":["mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.458}]}},{"model":"grok-4-6-xhigh","benchmark":"tau-bank/3-banking","score":0.474,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.03,"reasons":["cross-validated error ±5.0 pp"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":8,"inputs":[{"benchmark":"automationbench/1.0","score":0.67}]}},{"model":"gemini-3-5-flash-high","benchmark":"tau-bank/3-banking","score":0.502775,"source":"estimated","estimate":{"confidence":"low","error_pp":11.4,"reasons":["large cross-validated error ±11.4 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":13,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.501}]}},{"model":"deepseek-v4-pro-0813-max","benchmark":"tau-bank/3-banking","score":0.500653,"source":"estimated","estimate":{"confidence":"low","error_pp":11.4,"reasons":["large cross-validated error ±11.4 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":13,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.496}]}},{"model":"gpt-5-6-terra-max","benchmark":"tau-bank/3-banking","score":0.456079,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.62,"reasons":["mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.51}]}},{"model":"glm-5-2-max","benchmark":"tau-bank/3-banking","score":0.478348,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.62,"reasons":["mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.427}]}},{"model":"claude-opus-4-7-max","benchmark":"tau-bank/3-banking","score":0.467616,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.62,"reasons":["mapping explains only 34% of the variance (R²=0.34)"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":24,"inputs":[{"benchmark":"itbench/sre","score":0.467}]}},{"model":"minimax-m3","benchmark":"ifbench/1.0","score":0.829,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"ifbench/1.0","score":0.814,"source":"measured","estimate":null},{"model":"grok-4-3-medium","benchmark":"ifbench/1.0","score":0.833,"source":"measured","estimate":null},{"model":"grok-4-20-0309","benchmark":"ifbench/1.0","score":0.829,"source":"measured","estimate":null},{"model":"grok-4-3-high","benchmark":"ifbench/1.0","score":0.813,"source":"measured","estimate":null},{"model":"grok-4-20-0309-v2","benchmark":"ifbench/1.0","score":0.812,"source":"measured","estimate":null},{"model":"grok-4-3-low","benchmark":"ifbench/1.0","score":0.81,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"mmmu-pro/1.0","score":0.847924,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.262}]}},{"model":"gpt-6-astra-high","benchmark":"mmmu-pro/1.0","score":0.86,"source":"measured","estimate":null},{"model":"gpt-6-astra-medium","benchmark":"mmmu-pro/1.0","score":0.86388,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.304}]}},{"model":"qwen3-8-max-0902","benchmark":"mmmu-pro/1.0","score":0.83,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"mmmu-pro/1.0","score":0.87,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"mmmu-pro/1.0","score":0.86,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"mmmu-pro/1.0","score":0.81,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"mmmu-pro/1.0","score":0.849985,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.266}]}},{"model":"glm-5-3-flash","benchmark":"mmmu-pro/1.0","score":0.765592,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.154}]}},{"model":"glm-5-3-max","benchmark":"mmmu-pro/1.0","score":0.753208,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.112}]}},{"model":"qwen3-8-27b-xhigh","benchmark":"mmmu-pro/1.0","score":0.76,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"mmmu-pro/1.0","score":0.750981,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.6,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.074}]}},{"model":"minimax-m3","benchmark":"mmmu-pro/1.0","score":0.79,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"mmmu-pro/1.0","score":0.73,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"mmmu-pro/1.0","score":0.750795,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.6,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.05}]}},{"model":"muse-glimmer-high","benchmark":"mmmu-pro/1.0","score":0.74,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"mmmu-pro/1.0","score":0.84573,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.258}]}},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"mmmu-pro/1.0","score":0.88,"source":"measured","estimate":null},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"mmmu-pro/1.0","score":0.87,"source":"measured","estimate":null},{"model":"gpt-6-astra-xhigh","benchmark":"mmmu-pro/1.0","score":0.86,"source":"measured","estimate":null},{"model":"gemini-4-argon-high","benchmark":"mmmu-pro/1.0","score":0.816361,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.218}]}},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"mmmu-pro/1.0","score":0.86,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-max","benchmark":"mmmu-pro/1.0","score":0.86,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"mmmu-pro/1.0","score":0.83,"source":"measured","estimate":null},{"model":"mimo-v2-6-pro","benchmark":"mmmu-pro/1.0","score":0.792937,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.192}]}},{"model":"step-5-preview","benchmark":"mmmu-pro/1.0","score":0.76,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"mmmu-pro/1.0","score":0.77,"source":"measured","estimate":null},{"model":"mistral-large-4-preview","benchmark":"mmmu-pro/1.0","score":0.76,"source":"measured","estimate":null},{"model":"grok-4-7-xhigh","benchmark":"mmmu-pro/1.0","score":0.800115,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.2}]}},{"model":"gpt-6-luna-max","benchmark":"mmmu-pro/1.0","score":0.8,"source":"measured","estimate":null},{"model":"claude-opus-5-5-medium-with-fallback","benchmark":"mmmu-pro/1.0","score":0.86,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-xhigh","benchmark":"mmmu-pro/1.0","score":0.867018,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.318}]}},{"model":"gpt-6-1-sol-high","benchmark":"mmmu-pro/1.0","score":0.867403,"source":"estimated","estimate":{"confidence":"high","error_pp":2.6,"reasons":[],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":14,"inputs":[{"benchmark":"gdp-pdf/1.0","score":0.32}]}},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"analyst-agent/1.0","score":0.575,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"analyst-agent/1.0","score":0.4729,"source":"estimated","estimate":{"confidence":"low","error_pp":13.87,"reasons":["large cross-validated error ±13.9 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":17,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.933}]}},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"analyst-agent/1.0","score":0.460985,"source":"estimated","estimate":{"confidence":"low","error_pp":13.87,"reasons":["large cross-validated error ±13.9 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":17,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.93}]}},{"model":"gpt-6-astra-high","benchmark":"analyst-agent/1.0","score":0.533717,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.666}]}},{"model":"qwen3-8-max-0902","benchmark":"analyst-agent/1.0","score":0.45,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"analyst-agent/1.0","score":0.512,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"analyst-agent/1.0","score":0.477674,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.599}]}},{"model":"kimi-k3-max","benchmark":"analyst-agent/1.0","score":0.388,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"analyst-agent/1.0","score":0.455978,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.579}]}},{"model":"glm-5-3-flash","benchmark":"analyst-agent/1.0","score":0.13512,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.55,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":11,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.332}]}},{"model":"glm-5-3-max","benchmark":"analyst-agent/1.0","score":0.193601,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.55,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":11,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.364}]}},{"model":"qwen3-8-27b-xhigh","benchmark":"analyst-agent/1.0","score":0.344804,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.55,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":11,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.442}]}},{"model":"k2-horizon-375b-a23b","benchmark":"analyst-agent/1.0","score":0.191179,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.372}]}},{"model":"minimax-m3","benchmark":"analyst-agent/1.0","score":0.1,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"analyst-agent/1.0","score":0.238,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"analyst-agent/1.0","score":0.063,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"analyst-agent/1.0","score":0.162288,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.55,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":11,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.347}]}},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"analyst-agent/1.0","score":0.575,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"analyst-agent/1.0","score":0.563,"source":"measured","estimate":null},{"model":"gpt-6-astra-xhigh","benchmark":"analyst-agent/1.0","score":0.537592,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.672}]}},{"model":"gemini-4-argon-high","benchmark":"analyst-agent/1.0","score":0.582821,"source":"estimated","estimate":{"confidence":"low","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.775}]}},{"model":"gpt-6-1-sol-max","benchmark":"analyst-agent/1.0","score":0.5,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"analyst-agent/1.0","score":0.494269,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.616}]}},{"model":"mimo-v2-6-pro","benchmark":"analyst-agent/1.0","score":0.463837,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.586}]}},{"model":"step-5-preview","benchmark":"analyst-agent/1.0","score":0.35,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"analyst-agent/1.0","score":0.547688,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.689}]}},{"model":"mistral-large-4-preview","benchmark":"analyst-agent/1.0","score":0.477674,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.599}]}},{"model":"grok-4-7-xhigh","benchmark":"analyst-agent/1.0","score":0.526875,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.656}]}},{"model":"gpt-6-luna-max","benchmark":"analyst-agent/1.0","score":0.39641,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.532}]}},{"model":"gpt-5-6-sol-max","benchmark":"analyst-agent/1.0","score":0.475,"source":"measured","estimate":null},{"model":"claude-fable-5-with-fallback","benchmark":"analyst-agent/1.0","score":0.488,"source":"measured","estimate":null},{"model":"grok-4-6-high","benchmark":"analyst-agent/1.0","score":0.429624,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.55,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":11,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.483}]}},{"model":"gemini-3-7-flash-high","benchmark":"analyst-agent/1.0","score":0.6,"source":"measured","estimate":null},{"model":"gemini-3-7-flash-medium","benchmark":"analyst-agent/1.0","score":0.474608,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.55,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":11,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.504}]}},{"model":"gpt-6-1-sol-xhigh","benchmark":"analyst-agent/1.0","score":0.533717,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.666}]}},{"model":"qwen3-8-max","benchmark":"analyst-agent/1.0","score":0.547824,"source":"estimated","estimate":{"confidence":"low","error_pp":8.11,"reasons":["cross-validated error ±8.1 pp","extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":26,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.513}]}},{"model":"qwen3-8-2-4t-a95b","benchmark":"analyst-agent/1.0","score":0.519719,"source":"estimated","estimate":{"confidence":"low","error_pp":8.11,"reasons":["cross-validated error ±8.1 pp","extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":26,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.491}]}},{"model":"qwen3-8-27b-medium","benchmark":"analyst-agent/1.0","score":0.498001,"source":"estimated","estimate":{"confidence":"low","error_pp":8.11,"reasons":["cross-validated error ±8.1 pp","fitted on only 7 paired models"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":26,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.474}]}},{"model":"muse-spark-1-3-xhigh","benchmark":"analyst-agent/1.0","score":0.495445,"source":"estimated","estimate":{"confidence":"low","error_pp":8.11,"reasons":["cross-validated error ±8.1 pp","fitted on only 7 paired models"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":26,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.472}]}},{"model":"claude-opus-5-max","benchmark":"analyst-agent/1.0","score":0.538,"source":"measured","estimate":null},{"model":"gpt-5-5-xhigh","benchmark":"analyst-agent/1.0","score":0.5,"source":"measured","estimate":null},{"model":"grok-4-6-xhigh","benchmark":"analyst-agent/1.0","score":0.536319,"source":"estimated","estimate":{"confidence":"medium","error_pp":7.84,"reasons":["cross-validated error ±7.8 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":5,"inputs":[{"benchmark":"automationbench/1.0","score":0.67}]}},{"model":"gemini-3-5-flash-high","benchmark":"analyst-agent/1.0","score":0.468115,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.55,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":11,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.501}]}},{"model":"deepseek-v4-pro-0813-max","benchmark":"analyst-agent/1.0","score":0.457344,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.55,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":11,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.496}]}},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"harvey-lab/1.0","score":0.93,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"harvey-lab/1.0","score":0.933,"source":"measured","estimate":null},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"harvey-lab/1.0","score":0.93,"source":"measured","estimate":null},{"model":"gpt-6-astra-high","benchmark":"harvey-lab/1.0","score":0.93875,"source":"estimated","estimate":{"confidence":"high","error_pp":3.19,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":7,"inputs":[{"benchmark":"automationbench/1.0","score":0.666}]}},{"model":"qwen3-8-max-0902","benchmark":"harvey-lab/1.0","score":0.936,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"harvey-lab/1.0","score":0.936408,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.486}]}},{"model":"gemini-3-8-flash-high","benchmark":"harvey-lab/1.0","score":0.93831,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.525}]}},{"model":"kimi-k3-max","benchmark":"harvey-lab/1.0","score":0.946,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"harvey-lab/1.0","score":0.893238,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.332}]}},{"model":"glm-5-3-flash","benchmark":"harvey-lab/1.0","score":0.887783,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.332}]}},{"model":"glm-5-3-max","benchmark":"harvey-lab/1.0","score":0.911574,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.364}]}},{"model":"qwen3-8-27b-xhigh","benchmark":"harvey-lab/1.0","score":0.934969,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.442}]}},{"model":"k2-horizon-375b-a23b","benchmark":"harvey-lab/1.0","score":0.94175,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.09,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":29,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.342}]}},{"model":"minimax-m3","benchmark":"harvey-lab/1.0","score":0.884,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"harvey-lab/1.0","score":0.919179,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.38}]}},{"model":"nemotron-3-ultra","benchmark":"harvey-lab/1.0","score":0.817,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"harvey-lab/1.0","score":0.900687,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.347}]}},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"harvey-lab/1.0","score":0.931,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"harvey-lab/1.0","score":0.912,"source":"measured","estimate":null},{"model":"gpt-6-astra-xhigh","benchmark":"harvey-lab/1.0","score":0.939678,"source":"estimated","estimate":{"confidence":"high","error_pp":3.19,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":7,"inputs":[{"benchmark":"automationbench/1.0","score":0.672}]}},{"model":"gemini-4-argon-high","benchmark":"harvey-lab/1.0","score":0.955602,"source":"estimated","estimate":{"confidence":"medium","error_pp":3.19,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":7,"inputs":[{"benchmark":"automationbench/1.0","score":0.775}]}},{"model":"gpt-6-1-sol-max","benchmark":"harvey-lab/1.0","score":0.933725,"source":"estimated","estimate":{"confidence":"high","error_pp":1.8,"reasons":[],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":3,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.5}]}},{"model":"gpt-6-sol-max","benchmark":"harvey-lab/1.0","score":0.936886,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.494}]}},{"model":"mimo-v2-6-pro","benchmark":"harvey-lab/1.0","score":0.926382,"source":"estimated","estimate":{"confidence":"high","error_pp":3.19,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":7,"inputs":[{"benchmark":"automationbench/1.0","score":0.586}]}},{"model":"step-5-preview","benchmark":"harvey-lab/1.0","score":0.934,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"harvey-lab/1.0","score":0.93519,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.469}]}},{"model":"mistral-large-4-preview","benchmark":"harvey-lab/1.0","score":0.928392,"source":"estimated","estimate":{"confidence":"high","error_pp":3.19,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":7,"inputs":[{"benchmark":"automationbench/1.0","score":0.599}]}},{"model":"grok-4-7-xhigh","benchmark":"harvey-lab/1.0","score":0.929464,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.421}]}},{"model":"gpt-6-luna-max","benchmark":"harvey-lab/1.0","score":0.918033,"source":"estimated","estimate":{"confidence":"high","error_pp":3.19,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":7,"inputs":[{"benchmark":"automationbench/1.0","score":0.532}]}},{"model":"gpt-5-6-sol-max","benchmark":"harvey-lab/1.0","score":0.939396,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.562}]}},{"model":"claude-fable-5-with-fallback","benchmark":"harvey-lab/1.0","score":0.936,"source":"measured","estimate":null},{"model":"grok-4-6-high","benchmark":"harvey-lab/1.0","score":0.939481,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.483}]}},{"model":"gemini-3-7-flash-high","benchmark":"harvey-lab/1.0","score":0.936589,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.8,"reasons":["extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":3,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.6}]}},{"model":"gemini-3-7-flash-medium","benchmark":"harvey-lab/1.0","score":0.940937,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.504}]}},{"model":"gpt-6-1-sol-xhigh","benchmark":"harvey-lab/1.0","score":0.93875,"source":"estimated","estimate":{"confidence":"high","error_pp":3.19,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":7,"inputs":[{"benchmark":"automationbench/1.0","score":0.666}]}},{"model":"qwen3-8-max","benchmark":"harvey-lab/1.0","score":0.94175,"source":"estimated","estimate":{"confidence":"low","error_pp":2.09,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 6 paired models"],"extrapolated":true,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":29,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.513}]}},{"model":"qwen3-8-2-4t-a95b","benchmark":"harvey-lab/1.0","score":0.94175,"source":"estimated","estimate":{"confidence":"low","error_pp":2.09,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 6 paired models"],"extrapolated":true,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":29,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.491}]}},{"model":"qwen3-8-27b-medium","benchmark":"harvey-lab/1.0","score":0.94175,"source":"estimated","estimate":{"confidence":"medium","error_pp":2.09,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"logistic","method_name":"offset logistic","kind":"univariate","mapping_id":29,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.474}]}},{"model":"muse-spark-1-3-xhigh","benchmark":"harvey-lab/1.0","score":0.955,"source":"measured","estimate":null},{"model":"claude-opus-5-max","benchmark":"harvey-lab/1.0","score":0.935,"source":"measured","estimate":null},{"model":"gpt-5-5-xhigh","benchmark":"harvey-lab/1.0","score":0.934224,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.458}]}},{"model":"grok-4-6-xhigh","benchmark":"harvey-lab/1.0","score":0.939368,"source":"estimated","estimate":{"confidence":"high","error_pp":3.19,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":7,"inputs":[{"benchmark":"automationbench/1.0","score":0.67}]}},{"model":"gemini-3-5-flash-high","benchmark":"harvey-lab/1.0","score":0.940754,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.501}]}},{"model":"deepseek-v4-pro-0813-max","benchmark":"harvey-lab/1.0","score":0.940432,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.18,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":12,"inputs":[{"benchmark":"enterprise-ops-gym/1.0","score":0.496}]}},{"model":"gpt-5-6-terra-max","benchmark":"harvey-lab/1.0","score":0.937696,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.51}]}},{"model":"glm-5-2-max","benchmark":"harvey-lab/1.0","score":0.930436,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.427}]}},{"model":"claude-opus-4-7-max","benchmark":"harvey-lab/1.0","score":0.935026,"source":"estimated","estimate":{"confidence":"low","error_pp":1.62,"reasons":["fitted on only 5 paired models","mapping explains only 44% of the variance (R²=0.44)"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":23,"inputs":[{"benchmark":"itbench/sre","score":0.467}]}},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"hle/1.0","score":0.591,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"hle/1.0","score":0.587,"source":"measured","estimate":null},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"hle/1.0","score":0.559,"source":"measured","estimate":null},{"model":"gpt-6-astra-high","benchmark":"hle/1.0","score":0.504274,"source":"estimated","estimate":{"confidence":"high","error_pp":4.58,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":16,"inputs":[{"benchmark":"gpqa/diamond","score":0.949}]}},{"model":"qwen3-8-max-0902","benchmark":"hle/1.0","score":0.431,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"hle/1.0","score":0.547,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"hle/1.0","score":0.478,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"hle/1.0","score":0.469,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"hle/1.0","score":0.487,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"hle/1.0","score":0.399,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"hle/1.0","score":0.423,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"hle/1.0","score":0.339,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"hle/1.0","score":0.32,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"hle/1.0","score":0.39,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"hle/1.0","score":0.319,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"hle/1.0","score":0.284,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"hle/1.0","score":0.22,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"hle/1.0","score":0.55,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"hle/1.0","score":0.614,"source":"measured","estimate":null},{"model":"claude-opus-5-5-xhigh-with-fallback","benchmark":"hle/1.0","score":0.575,"source":"measured","estimate":null},{"model":"gpt-6-astra-xhigh","benchmark":"hle/1.0","score":0.568346,"source":"estimated","estimate":{"confidence":"high","error_pp":3.39,"reasons":[],"extrapolated":false,"method":"linear_mv","method_name":"ridge regression (multivariate)","kind":"multivariate","mapping_id":null,"inputs":[{"benchmark":"critpt/1.0","score":0.314},{"benchmark":"gpqa/diamond","score":0.963}]}},{"model":"gemini-4-argon-high","benchmark":"hle/1.0","score":0.571,"source":"measured","estimate":null},{"model":"claude-opus-5-5-high-with-fallback","benchmark":"hle/1.0","score":0.556,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-max","benchmark":"hle/1.0","score":0.529,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"hle/1.0","score":0.479,"source":"measured","estimate":null},{"model":"mimo-v2-6-pro","benchmark":"hle/1.0","score":0.494,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"hle/1.0","score":0.465,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"hle/1.0","score":0.392,"source":"measured","estimate":null},{"model":"mistral-large-4-preview","benchmark":"hle/1.0","score":0.35,"source":"measured","estimate":null},{"model":"grok-4-7-xhigh","benchmark":"hle/1.0","score":0.431,"source":"measured","estimate":null},{"model":"gpt-6-luna-max","benchmark":"hle/1.0","score":0.385,"source":"measured","estimate":null},{"model":"gpt-5-6-sol-max","benchmark":"hle/1.0","score":0.544584,"source":"estimated","estimate":{"confidence":"high","error_pp":3.39,"reasons":[],"extrapolated":false,"method":"linear_mv","method_name":"ridge regression (multivariate)","kind":"multivariate","mapping_id":null,"inputs":[{"benchmark":"critpt/1.0","score":0.323},{"benchmark":"gpqa/diamond","score":0.941}]}},{"model":"claude-fable-5-with-fallback","benchmark":"hle/1.0","score":0.555,"source":"measured","estimate":null},{"model":"grok-4-6-high","benchmark":"hle/1.0","score":0.504274,"source":"estimated","estimate":{"confidence":"high","error_pp":4.58,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":16,"inputs":[{"benchmark":"gpqa/diamond","score":0.949}]}},{"model":"gemini-3-7-flash-high","benchmark":"hle/1.0","score":0.49393,"source":"estimated","estimate":{"confidence":"high","error_pp":4.58,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":16,"inputs":[{"benchmark":"gpqa/diamond","score":0.945}]}},{"model":"gpt-6-1-sol-xhigh","benchmark":"hle/1.0","score":0.560798,"source":"estimated","estimate":{"confidence":"high","error_pp":4.07,"reasons":[],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":10,"inputs":[{"benchmark":"critpt/1.0","score":0.317}]}},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"automationbench/1.0","score":0.594,"source":"measured","estimate":null},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"automationbench/1.0","score":0.595936,"source":"estimated","estimate":{"confidence":"low","error_pp":13.03,"reasons":["large cross-validated error ±13.0 pp"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":18,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.933}]}},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"automationbench/1.0","score":0.581065,"source":"estimated","estimate":{"confidence":"low","error_pp":13.03,"reasons":["large cross-validated error ±13.0 pp"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":18,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.93}]}},{"model":"gpt-6-astra-high","benchmark":"automationbench/1.0","score":0.666,"source":"measured","estimate":null},{"model":"qwen3-8-max-0902","benchmark":"automationbench/1.0","score":0.562,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"automationbench/1.0","score":0.685,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"automationbench/1.0","score":0.599,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"automationbench/1.0","score":0.583,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"automationbench/1.0","score":0.579,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"automationbench/1.0","score":0.604,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"automationbench/1.0","score":0.622,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"automationbench/1.0","score":0.482,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"automationbench/1.0","score":0.372,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"automationbench/1.0","score":0.213,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"automationbench/1.0","score":0.05,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"automationbench/1.0","score":0.03,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"automationbench/1.0","score":0.068,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"automationbench/1.0","score":0.718,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"automationbench/1.0","score":0.695,"source":"measured","estimate":null},{"model":"gpt-6-astra-xhigh","benchmark":"automationbench/1.0","score":0.672,"source":"measured","estimate":null},{"model":"gemini-4-argon-high","benchmark":"automationbench/1.0","score":0.775,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-max","benchmark":"automationbench/1.0","score":0.649,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"automationbench/1.0","score":0.616,"source":"measured","estimate":null},{"model":"mimo-v2-6-pro","benchmark":"automationbench/1.0","score":0.586,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"automationbench/1.0","score":0.51,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"automationbench/1.0","score":0.689,"source":"measured","estimate":null},{"model":"mistral-large-4-preview","benchmark":"automationbench/1.0","score":0.599,"source":"measured","estimate":null},{"model":"grok-4-7-xhigh","benchmark":"automationbench/1.0","score":0.656,"source":"measured","estimate":null},{"model":"gpt-6-luna-max","benchmark":"automationbench/1.0","score":0.532,"source":"measured","estimate":null},{"model":"gpt-5-6-sol-max","benchmark":"automationbench/1.0","score":0.571534,"source":"estimated","estimate":{"confidence":"low","error_pp":11.19,"reasons":["large cross-validated error ±11.2 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":1,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.475}]}},{"model":"claude-fable-5-with-fallback","benchmark":"automationbench/1.0","score":0.592224,"source":"estimated","estimate":{"confidence":"low","error_pp":11.19,"reasons":["large cross-validated error ±11.2 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":1,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.488}]}},{"model":"grok-4-6-high","benchmark":"automationbench/1.0","score":0.667,"source":"measured","estimate":null},{"model":"gemini-3-7-flash-high","benchmark":"automationbench/1.0","score":0.786396,"source":"estimated","estimate":{"confidence":"low","error_pp":11.19,"reasons":["large cross-validated error ±11.2 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":1,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.6}]}},{"model":"gpt-6-1-sol-xhigh","benchmark":"automationbench/1.0","score":0.666,"source":"measured","estimate":null},{"model":"qwen3-8-max","benchmark":"automationbench/1.0","score":0.628077,"source":"estimated","estimate":{"confidence":"low","error_pp":10.6,"reasons":["large cross-validated error ±10.6 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":27,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.513}]}},{"model":"qwen3-8-2-4t-a95b","benchmark":"automationbench/1.0","score":0.611675,"source":"estimated","estimate":{"confidence":"low","error_pp":10.6,"reasons":["large cross-validated error ±10.6 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":27,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.491}]}},{"model":"qwen3-8-27b-medium","benchmark":"automationbench/1.0","score":0.595927,"source":"estimated","estimate":{"confidence":"low","error_pp":10.6,"reasons":["large cross-validated error ±10.6 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":27,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.474}]}},{"model":"muse-spark-1-3-xhigh","benchmark":"automationbench/1.0","score":0.593873,"source":"estimated","estimate":{"confidence":"low","error_pp":10.6,"reasons":["large cross-validated error ±10.6 pp"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":27,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.472}]}},{"model":"claude-opus-5-max","benchmark":"automationbench/1.0","score":0.675232,"source":"estimated","estimate":{"confidence":"low","error_pp":11.19,"reasons":["large cross-validated error ±11.2 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":1,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.538}]}},{"model":"gpt-5-5-xhigh","benchmark":"automationbench/1.0","score":0.611641,"source":"estimated","estimate":{"confidence":"low","error_pp":11.19,"reasons":["large cross-validated error ±11.2 pp"],"extrapolated":false,"method":"mm_offset_inv","method_name":"inverse Michaelis–Menten","kind":"univariate","mapping_id":1,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.5}]}},{"model":"grok-4-6-xhigh","benchmark":"automationbench/1.0","score":0.67,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"enterprise-ops-gym/1.0","score":0.554048,"source":"estimated","estimate":{"confidence":"low","error_pp":1.79,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.575}]}},{"model":"claude-fable-5-1-xhigh-with-fallback","benchmark":"enterprise-ops-gym/1.0","score":0.460382,"source":"estimated","estimate":{"confidence":"medium","error_pp":4.4,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":19,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.933}]}},{"model":"claude-fable-5-1-high-with-fallback","benchmark":"enterprise-ops-gym/1.0","score":0.455049,"source":"estimated","estimate":{"confidence":"medium","error_pp":4.4,"reasons":["fitted on only 6 paired models"],"extrapolated":false,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":19,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.93}]}},{"model":"gpt-6-astra-high","benchmark":"enterprise-ops-gym/1.0","score":0.423489,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.666}]}},{"model":"qwen3-8-max-0902","benchmark":"enterprise-ops-gym/1.0","score":0.476,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"enterprise-ops-gym/1.0","score":0.522472,"source":"estimated","estimate":{"confidence":"low","error_pp":1.79,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.512}]}},{"model":"gemini-3-8-flash-high","benchmark":"enterprise-ops-gym/1.0","score":0.418832,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.31,"reasons":["cross-validated error ±5.3 pp"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":28,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.449}]}},{"model":"kimi-k3-max","benchmark":"enterprise-ops-gym/1.0","score":0.453,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"enterprise-ops-gym/1.0","score":0.427846,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.31,"reasons":["cross-validated error ±5.3 pp"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":28,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.505}]}},{"model":"glm-5-3-flash","benchmark":"enterprise-ops-gym/1.0","score":0.332,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"enterprise-ops-gym/1.0","score":0.364,"source":"measured","estimate":null},{"model":"qwen3-8-27b-xhigh","benchmark":"enterprise-ops-gym/1.0","score":0.442,"source":"measured","estimate":null},{"model":"k2-horizon-375b-a23b","benchmark":"enterprise-ops-gym/1.0","score":0.395335,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.31,"reasons":["cross-validated error ±5.3 pp"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":28,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.342}]}},{"model":"minimax-m3","benchmark":"enterprise-ops-gym/1.0","score":0.321,"source":"measured","estimate":null},{"model":"inkling-xhigh","benchmark":"enterprise-ops-gym/1.0","score":0.38,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"enterprise-ops-gym/1.0","score":0.289,"source":"measured","estimate":null},{"model":"muse-glimmer-high","benchmark":"enterprise-ops-gym/1.0","score":0.347,"source":"measured","estimate":null},{"model":"claude-sonnet-5-5-max-with-fallback","benchmark":"enterprise-ops-gym/1.0","score":0.554048,"source":"estimated","estimate":{"confidence":"low","error_pp":1.79,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.575}]}},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"enterprise-ops-gym/1.0","score":0.548034,"source":"estimated","estimate":{"confidence":"low","error_pp":1.79,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.563}]}},{"model":"gpt-6-astra-xhigh","benchmark":"enterprise-ops-gym/1.0","score":0.423565,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","extrapolated beyond the scores the mapping was fitted on","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":true,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.672}]}},{"model":"gemini-4-argon-high","benchmark":"enterprise-ops-gym/1.0","score":0.42468,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","extrapolated beyond the scores the mapping was fitted on","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":true,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.775}]}},{"model":"gpt-6-1-sol-max","benchmark":"enterprise-ops-gym/1.0","score":0.516458,"source":"estimated","estimate":{"confidence":"low","error_pp":1.79,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.5}]}},{"model":"gpt-6-sol-max","benchmark":"enterprise-ops-gym/1.0","score":0.422805,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.616}]}},{"model":"mimo-v2-6-pro","benchmark":"enterprise-ops-gym/1.0","score":0.42234,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.586}]}},{"model":"step-5-preview","benchmark":"enterprise-ops-gym/1.0","score":0.472,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"enterprise-ops-gym/1.0","score":0.423771,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","extrapolated beyond the scores the mapping was fitted on","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":true,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.689}]}},{"model":"mistral-large-4-preview","benchmark":"enterprise-ops-gym/1.0","score":0.422547,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.599}]}},{"model":"grok-4-7-xhigh","benchmark":"enterprise-ops-gym/1.0","score":0.423361,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.656}]}},{"model":"gpt-6-luna-max","benchmark":"enterprise-ops-gym/1.0","score":0.421373,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.532}]}},{"model":"gpt-5-6-sol-max","benchmark":"enterprise-ops-gym/1.0","score":0.503928,"source":"estimated","estimate":{"confidence":"medium","error_pp":1.79,"reasons":["fitted on only 7 paired models"],"extrapolated":false,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.475}]}},{"model":"claude-fable-5-with-fallback","benchmark":"enterprise-ops-gym/1.0","score":0.511,"source":"measured","estimate":null},{"model":"grok-4-6-high","benchmark":"enterprise-ops-gym/1.0","score":0.483,"source":"measured","estimate":null},{"model":"gemini-3-7-flash-high","benchmark":"enterprise-ops-gym/1.0","score":0.566578,"source":"estimated","estimate":{"confidence":"low","error_pp":1.79,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.6}]}},{"model":"gemini-3-7-flash-medium","benchmark":"enterprise-ops-gym/1.0","score":0.504,"source":"measured","estimate":null},{"model":"gpt-6-1-sol-xhigh","benchmark":"enterprise-ops-gym/1.0","score":0.423489,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.666}]}},{"model":"qwen3-8-max","benchmark":"enterprise-ops-gym/1.0","score":0.429,"source":"estimated","estimate":{"confidence":"low","error_pp":5.31,"reasons":["cross-validated error ±5.3 pp","extrapolated beyond the scores the mapping was fitted on"],"extrapolated":true,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":28,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.513}]}},{"model":"qwen3-8-2-4t-a95b","benchmark":"enterprise-ops-gym/1.0","score":0.425751,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.31,"reasons":["cross-validated error ±5.3 pp"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":28,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.491}]}},{"model":"qwen3-8-27b-medium","benchmark":"enterprise-ops-gym/1.0","score":0.423072,"source":"estimated","estimate":{"confidence":"medium","error_pp":5.31,"reasons":["cross-validated error ±5.3 pp"],"extrapolated":false,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":28,"inputs":[{"benchmark":"tau-bank/3-banking","score":0.474}]}},{"model":"muse-spark-1-3-xhigh","benchmark":"enterprise-ops-gym/1.0","score":0.499698,"source":"estimated","estimate":{"confidence":"low","error_pp":4.4,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 6 paired models"],"extrapolated":true,"method":"hill","method_name":"Hill","kind":"univariate","mapping_id":19,"inputs":[{"benchmark":"harvey-lab/1.0","score":0.955}]}},{"model":"claude-opus-5-max","benchmark":"enterprise-ops-gym/1.0","score":0.535504,"source":"estimated","estimate":{"confidence":"low","error_pp":1.79,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.538}]}},{"model":"gpt-5-5-xhigh","benchmark":"enterprise-ops-gym/1.0","score":0.516458,"source":"estimated","estimate":{"confidence":"low","error_pp":1.79,"reasons":["extrapolated beyond the scores the mapping was fitted on","fitted on only 7 paired models"],"extrapolated":true,"method":"linear","method_name":"linear","kind":"univariate","mapping_id":2,"inputs":[{"benchmark":"analyst-agent/1.0","score":0.5}]}},{"model":"grok-4-6-xhigh","benchmark":"enterprise-ops-gym/1.0","score":0.42354,"source":"estimated","estimate":{"confidence":"low","error_pp":6.19,"reasons":["cross-validated error ±6.2 pp","extrapolated beyond the scores the mapping was fitted on","mapping explains only 37% of the variance (R²=0.37)"],"extrapolated":true,"method":"mm","method_name":"Michaelis–Menten","kind":"univariate","mapping_id":6,"inputs":[{"benchmark":"automationbench/1.0","score":0.67}]}},{"model":"gemini-3-5-flash-high","benchmark":"enterprise-ops-gym/1.0","score":0.501,"source":"measured","estimate":null},{"model":"deepseek-v4-pro-0813-max","benchmark":"enterprise-ops-gym/1.0","score":0.496,"source":"measured","estimate":null},{"model":"qwen3-8-max-0902","benchmark":"apex-agents/1.0","score":0.424,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"apex-agents/1.0","score":0.413,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"apex-agents/1.0","score":0.38,"source":"measured","estimate":null},{"model":"gpt-5-5-xhigh","benchmark":"apex-agents/1.0","score":0.377,"source":"measured","estimate":null},{"model":"gemini-3-5-flash-high","benchmark":"apex-agents/1.0","score":0.471,"source":"measured","estimate":null},{"model":"gpt-5-6-terra-max","benchmark":"apex-agents/1.0","score":0.389,"source":"measured","estimate":null},{"model":"gpt-5-6-luna-max","benchmark":"apex-agents/1.0","score":0.358,"source":"measured","estimate":null},{"model":"glm-5-2-max","benchmark":"apex-agents/1.0","score":0.337,"source":"measured","estimate":null},{"model":"claude-fable-5-1-max-with-fallback","benchmark":"itbench/sre","score":0.495,"source":"measured","estimate":null},{"model":"qwen3-8-max-0902","benchmark":"itbench/sre","score":0.403,"source":"measured","estimate":null},{"model":"gpt-6-astra-max","benchmark":"itbench/sre","score":0.486,"source":"measured","estimate":null},{"model":"gemini-3-8-flash-high","benchmark":"itbench/sre","score":0.525,"source":"measured","estimate":null},{"model":"kimi-k3-max","benchmark":"itbench/sre","score":0.477,"source":"measured","estimate":null},{"model":"muse-spark-1-3-max","benchmark":"itbench/sre","score":0.332,"source":"measured","estimate":null},{"model":"glm-5-3-flash","benchmark":"itbench/sre","score":0.512,"source":"measured","estimate":null},{"model":"glm-5-3-max","benchmark":"itbench/sre","score":0.461,"source":"measured","estimate":null},{"model":"claude-opus-5-5-max-with-fallback","benchmark":"itbench/sre","score":0.382,"source":"measured","estimate":null},{"model":"gpt-6-sol-max","benchmark":"itbench/sre","score":0.494,"source":"measured","estimate":null},{"model":"step-5-preview","benchmark":"itbench/sre","score":0.556,"source":"measured","estimate":null},{"model":"deepseek-v4-1-flash-max","benchmark":"itbench/sre","score":0.469,"source":"measured","estimate":null},{"model":"grok-4-7-xhigh","benchmark":"itbench/sre","score":0.421,"source":"measured","estimate":null},{"model":"gpt-5-6-sol-max","benchmark":"itbench/sre","score":0.562,"source":"measured","estimate":null},{"model":"gpt-5-5-xhigh","benchmark":"itbench/sre","score":0.458,"source":"measured","estimate":null},{"model":"gpt-5-6-terra-max","benchmark":"itbench/sre","score":0.51,"source":"measured","estimate":null},{"model":"glm-5-2-max","benchmark":"itbench/sre","score":0.427,"source":"measured","estimate":null},{"model":"claude-opus-4-7-max","benchmark":"itbench/sre","score":0.467,"source":"measured","estimate":null},{"model":"minimax-m3","benchmark":"tau-bench/2-telecom","score":0.889,"source":"measured","estimate":null},{"model":"nemotron-3-ultra","benchmark":"tau-bench/2-telecom","score":0.833,"source":"measured","estimate":null},{"model":"claude-fable-5-with-fallback","benchmark":"tau-bench/2-telecom","score":0.985,"source":"measured","estimate":null},{"model":"glm-5-2-max","benchmark":"tau-bench/2-telecom","score":0.991,"source":"measured","estimate":null},{"model":"jt-35b-flash","benchmark":"tau-bench/2-telecom","score":0.991,"source":"measured","estimate":null},{"model":"glm-4-7-flash","benchmark":"tau-bench/2-telecom","score":0.988,"source":"measured","estimate":null},{"model":"glm-5-turbo","benchmark":"tau-bench/2-telecom","score":0.985,"source":"measured","estimate":null}]}