Calibration

CritPt → C-Eval

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

C-Eval is estimated from CritPt with a linear curve fitted on 5 models measured on both: y = 0.7851·x + 0.9112, R² = 0.40, cross-validated error 1.4 pp. It is used for 1 estimate.

Estimated model	CritPt	C-Eval	Source
Gemini 3 Pro Deep Think	25.7%	100.0%	estimated ± 1.4 pp, low confidence