benchgap
Calibration

PhysicianBench → HealthBench (raw, Artificial Analysis)

HealthBench (raw, Artificial Analysis) is estimated from PhysicianBench with a Michaelis–Menten + offset curve fitted on 5 models measured on both: y = 0.0000 + 0.8920·x / (0.26551 + x), R² = 0.47, cross-validated error 4.3 pp. It is used for 0 estimates.

Fitted on

ModelPhysicianBenchHealthBench (raw, Artificial Analysis)
Claude Fable 5.161.0%63.9%
Claude Fable 5.264.4%58.8%
Claude Opus 5.568.4%64.3%
Claude Sonnet 5.563.2%65.9%
GPT-6 Astra48.0%56.9%