Calibration
HealthBench Professional → HealthBench (length-adjusted)
HealthBench (length-adjusted) is estimated from HealthBench Professional with a inverse Michaelis–Menten curve fitted on 7 models measured on both: y = 1.24795·(x − 0.0000) / (0.0000 + 2.0000 − x), R² = 0.74, cross-validated error 2.2 pp. It is used for 3 estimates.
| Estimated model | HealthBench Professional | HealthBench (length-adjusted) | Source |
|---|---|---|---|
| Grok 4.7 | 56.7% | 49.4% | estimated ± 2.2 pp, low confidence |
| Ling 3.1 Flash | 65.4% | 60.6% | estimated ± 2.2 pp, medium confidence |
| Muse Spark 1.1 | 59.3% | 52.6% | estimated ± 2.2 pp, low confidence |