Calibration

HLE → CritPt

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

CritPt is estimated from HLE with a offset logistic curve fitted on 51 models measured on both: y = 0.0000 + (1.2000 − 0.0000) / (1 + exp(−5.42·(x − 0.8822))), R² = 0.54, cross-validated error 6.5 pp. It is used for 1 estimate.

Estimated model	HLE	CritPt	Source
Agents-A1	47.6%	12.0%	estimated ± 6.5 pp, medium confidence