Calibration
ARC-AGI-3 → AA-GPQA Diamond
AA-GPQA Diamond is estimated from ARC-AGI-3 with a Michaelis–Menten curve fitted on 13 models measured on both: y = 0.9442·x / (0.00006 + x), R² = 0.54, cross-validated error 1.2 pp. It is used for 4 estimates.
| Estimated model | ARC-AGI-3 | AA-GPQA Diamond | Source |
|---|---|---|---|
| GPT-6.1 Sol | 52.7% | 94.4% | estimated ± 1.2 pp, high confidence |
| GPT-6 Luna | 0.1% | 89.3% | estimated ± 1.2 pp, medium confidence |
| GPT-6 Sol | 4.6% | 94.3% | estimated ± 1.2 pp, high confidence |
| Grok 4.20 | 0.1% | 88.7% | estimated ± 1.2 pp, medium confidence |