Calibration

HLE → AA-HLE

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

AA-HLE is estimated from HLE with a linear curve fitted on 48 models measured on both: y = 0.5557·x + 0.1123, R² = 0.51, cross-validated error 8.4 pp. It is used for 1 estimate.

Estimated model	HLE	AA-HLE	Source
Agents-A1	47.6%	37.7%	estimated ± 8.4 pp, medium confidence