benchgap
Calibration

OSWorld 2.0 → Gert Labs

Gert Labs is estimated from OSWorld 2.0 with a Michaelis–Menten curve fitted on 5 models measured on both: y = 0.7869·x / (0.01843 + x), R² = 0.82, cross-validated error 3.7 pp. It is used for 9 estimates.

Estimated modelOSWorld 2.0Gert LabsSource
Claude Opus 4.7 (Adaptive)18.2%71.5%estimated ± 3.7 pp, medium confidence
Claude Opus 5.548.7%75.8%estimated ± 3.7 pp, low confidence
Gemini 4 Argon69.2%76.7%estimated ± 3.7 pp, low confidence
GPT-6 Sol60.5%76.4%estimated ± 3.7 pp, low confidence
MiniMax M34.6%56.2%estimated ± 3.7 pp, medium confidence
Muse Spark 1.114.2%69.7%estimated ± 3.7 pp, medium confidence
Qwen3.7 Plus2.8%47.5%estimated ± 3.7 pp, low confidence
Qwen3.8-Flash-Next19.4%71.9%estimated ± 3.7 pp, medium confidence
Qwen3.8 Max19.4%71.9%estimated ± 3.7 pp, medium confidence