benchgap
Calibration

Toolathlon → AA Agentic Index

AA Agentic Index is estimated from Toolathlon with a offset logistic curve fitted on 15 models measured on both: y = 0.1655 + (0.3883 − 0.1655) / (1 + exp(−198.58·(x − 0.4691))), R² = 0.58, cross-validated error 10.4 pp. It is used for 0 estimates.

Fitted on

ModelToolathlonAA Agentic Index
Claude Opus 4.859.9%42.6%
DeepSeek V4 Flash 073147.8%41.7%
DeepSeek V4 Pro 081351.8%49.6%
Gemini 3.5 Flash56.5%27.3%
GLM-5.248.2%39.4%
GPT-5.4 mini42.9%19.7%
GPT-5.4 nano35.5%17.7%
GPT-5.555.6%37.3%
GPT-5.6 Luna53.4%42.7%
GPT-5.6 Sol58.0%50.5%
GPT-5.6 Terra53.1%43.7%
Kimi K2.650.0%22.1%
MiniMax M2.746.3%16.8%
Muse Spark 1.175.6%27.5%
Qwen3.6-35B-A3B26.9%15.0%