Calibration

HLE-Verified → LABBench2

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

LABBench2 is estimated from HLE-Verified with a offset logistic curve fitted on 6 models measured on both: y = 0.8080 + (1.2000 − 0.8080) / (1 + exp(−158.99·(x − 0.5610))), R² = 0.82, cross-validated error 1.8 pp. It is used for 0 estimates.