benchgap
Calibration

Claw-Eval → AA Agentic Index

AA Agentic Index is estimated from Claw-Eval with a inverse Michaelis–Menten curve fitted on 11 models measured on both: y = 0.43628·(x − 0.0000) / (0.0000 + 2.0000 − x), R² = 0.58, cross-validated error 5.1 pp. It is used for 18 estimates.

Estimated modelClaw-EvalAA Agentic IndexSource
Claude Opus 4.559.6%18.5%estimated ± 5.1 pp, medium confidence
Claude Opus 4.670.4%23.7%estimated ± 5.1 pp, medium confidence
Claude Sonnet 4.667.8%22.4%estimated ± 5.1 pp, medium confidence
DeepSeek V3.240.2%11.0%estimated ± 5.1 pp, medium confidence
Gemini 3 Flash49.2%14.2%estimated ± 5.1 pp, medium confidence
GLM-557.7%17.7%estimated ± 5.1 pp, medium confidence
GLM-5-Turbo55.8%16.9%estimated ± 5.1 pp, medium confidence
GLM-5V-Turbo53.8%16.1%estimated ± 5.1 pp, medium confidence
K-EXAONE 2.077.7%27.7%estimated ± 5.1 pp, low confidence
LLaDA2.2-flash64.2%20.6%estimated ± 5.1 pp, medium confidence
LLaDA2.2-mini57.2%17.5%estimated ± 5.1 pp, medium confidence
MiMo-V2.562.3%19.7%estimated ± 5.1 pp, medium confidence
MiMo-V2-Omni45.2%12.7%estimated ± 5.1 pp, medium confidence
MiMo-V2-Pro57.8%17.7%estimated ± 5.1 pp, medium confidence
Ornith-1.0-35B69.8%23.4%estimated ± 5.1 pp, medium confidence
Ornith-1.0-397B77.1%27.4%estimated ± 5.1 pp, low confidence
Ornith-1.0-9B63.1%20.1%estimated ± 5.1 pp, medium confidence
Qwen3.5 397B56.8%17.3%estimated ± 5.1 pp, medium confidence