benchgap
Calibration

Toolathlon → OSWorld-Verified

OSWorld-Verified is estimated from Toolathlon with a Michaelis–Menten curve fitted on 11 models measured on both: y = 1.3601·x / (0.44231 + x), R² = 0.57, cross-validated error 10.2 pp. It is used for 0 estimates.

Fitted on

ModelToolathlonOSWorld-Verified
Claude Opus 4.543.5%66.3%
Claude Opus 4.859.9%83.4%
Gemini 3.5 Flash56.5%78.4%
GPT-5.454.6%75.0%
GPT-5.4 mini42.9%72.1%
GPT-5.4 nano35.5%39.0%
GPT-5.555.6%78.7%
Kimi K2.650.0%73.1%
Kimi K2.527.8%63.3%
Muse Spark 1.175.6%80.8%
Qwen3.5 397B36.3%62.2%