Calibration

LABBench2 → HLE-Verified

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

HLE-Verified is estimated from LABBench2 with a offset logistic curve fitted on 6 models measured on both: y = 0.0000 + (0.5483 − 0.0000) / (1 + exp(−200.00·(x − 0.7996))), R² = 1.00, cross-validated error 1.2 pp. It is used for 1 estimate.

Estimated model	LABBench2	HLE-Verified	Source
Gemini 4 Argon	88.8%	54.8%	estimated ± 1.2 pp, low confidence