Calibration

HealthBench (length-adjusted) → HealthBench Professional (raw)

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

HealthBench Professional (raw) is estimated from HealthBench (length-adjusted) with a Hill curve fitted on 7 models measured on both: y = 0.0000 + (0.9104 − 0.0000)·x^6.00 / (0.47785^6.00 + x^6.00), R² = 0.84, cross-validated error 4.4 pp. It is used for 0 estimates.