Calibration
HealthBench (length-adjusted) → ARC-AGI-1
ARC-AGI-1 is estimated from HealthBench (length-adjusted) with a Michaelis–Menten curve fitted on 6 models measured on both: y = 2.0000·x / (0.62861 + x), R² = 0.36, cross-validated error 4.4 pp. It is used for 1 estimate.
| Estimated model | HealthBench (length-adjusted) | ARC-AGI-1 | Source |
|---|---|---|---|
| GPT-6.1 Terra | 55.1% | 93.4% | estimated ± 4.4 pp, low confidence |
Fitted on
| Model | HealthBench (length-adjusted) | ARC-AGI-1 |
|---|---|---|
| Claude Opus 5 | 57.8% | 97.5% |
| Claude Opus 5.5 | 60.6% | 97.5% |
| GPT-6.1 Sol | 58.5% | 96.5% |
| GPT-6 Astra | 58.3% | 97.5% |
| GPT-6 Luna | 54.5% | 86.7% |
| GPT-6 Sol | 53.2% | 95.5% |