Calibration

CritPt → MMLU-Pro

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

MMLU-Pro is estimated from CritPt with a Michaelis–Menten + offset curve fitted on 36 models measured on both: y = 0.7635 + 0.1249·x / (0.00774 + x), R² = 0.39, cross-validated error 5.8 pp. It is used for 1 estimate.

Estimated model	CritPt	MMLU-Pro	Source
Gemini 3 Pro Deep Think	25.7%	88.5%	estimated ± 5.8 pp, low confidence