Calibration

ERQA → ZeroBench

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

ZeroBench is estimated from ERQA with a Hill curve fitted on 6 models measured on both: y = 1.0000 + (0.0246 − 1.0000)·x^6.00 / (0.57447^6.00 + x^6.00), R² = 0.46, cross-validated error 9.3 pp. It is used for 1 estimate.

Estimated model	ERQA	ZeroBench	Source
Claude Opus 4.6	51.6%	66.4%	estimated ± 9.3 pp, low confidence