Calibration
Vals GPQA Diamond → HealthBench Professional
HealthBench Professional is estimated from Vals GPQA Diamond with a Michaelis–Menten + offset curve fitted on 5 models measured on both: y = 0.0000 + 2.0000·x / (2.23116 + x), R² = 0.37, cross-validated error 1.7 pp. It is used for 0 estimates.
Fitted on
| Model | Vals GPQA Diamond | HealthBench Professional |
|---|---|---|
| Claude Opus 5 | 93.4% | 59.8% |
| GPT-5.6 Luna | 91.7% | 55.7% |
| GPT-5.6 Sol | 95.2% | 60.5% |
| GPT-5.6 Terra | 90.9% | 57.7% |
| Muse Spark 1.1 | 91.2% | 59.3% |