benchgap
Calibration

Toolathlon → APEX-Agents-AA

APEX-Agents-AA is estimated from Toolathlon with a Hill curve fitted on 15 models measured on both: y = 0.1361 + (1.2000 − 0.1361)·x^6.00 / (0.68010^6.00 + x^6.00), R² = 0.61, cross-validated error 7.6 pp. It is used for 0 estimates.

Fitted on

ModelToolathlonAPEX-Agents-AA
DeepSeek V4 Pro 081351.8%24.3%
Gemini 3.5 Flash56.5%47.1%
GLM-538.0%14.5%
GLM-5.248.2%33.7%
GPT-5.454.6%33.3%
GPT-5.4 mini42.9%28.2%
GPT-5.4 nano35.5%24.9%
GPT-5.555.6%37.7%
GPT-5.6 Luna53.4%35.8%
GPT-5.6 Terra53.1%38.9%
Kimi K2.650.0%28.5%
Kimi K2.527.8%11.5%
MiniMax M2.746.3%10.6%
Qwen3.5 397B36.3%15.3%
Step 3.7 Flash49.5%14.8%