Calibration

AA EnterpriseOps-Gym → OSWorld-Verified

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

OSWorld-Verified is estimated from AA EnterpriseOps-Gym with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.3782·x / (0.33132 + x), R² = 0.73, cross-validated error 6.2 pp. It is used for 1 estimate.

Estimated model	AA EnterpriseOps-Gym	OSWorld-Verified	Source
GLM-5.3-Flash	33.2%	69.0%	estimated ± 6.2 pp, low confidence