Calibration

CritPt → Humanity's Last Exam

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

Humanity's Last Exam is estimated from CritPt with a linear curve fitted on 27 models measured on both: y = 0.9277·x + 0.2667, R² = 0.87, cross-validated error 4.1 pp. It is used for 1 estimate.

Estimated model	CritPt	Humanity's Last Exam	Source
GPT-6.1 Sol (xhigh)	31.7%	56.1%	estimated ± 4.1 pp, high confidence