Calibration
ARC-AGI-2 → AA-Omniscience Accuracy
AA-Omniscience Accuracy is estimated from ARC-AGI-2 with a linear curve fitted on 34 models measured on both: y = 0.3275·x + 0.2858, R² = 0.62, cross-validated error 6.2 pp. It is used for 6 estimates.
| Estimated model | ARC-AGI-2 | AA-Omniscience Accuracy | Source |
|---|---|---|---|
| Claude Sonnet 4.5 | 13.6% | 33.0% | estimated ± 6.2 pp, medium confidence |
| Claude Sonnet 4.5 Thinking | 13.6% | 33.0% | estimated ± 6.2 pp, medium confidence |
| dots3-note Preview | 81.4% | 55.2% | estimated ± 6.2 pp, medium confidence |
| Gemini 3 Pro Deep Think | 45.1% | 43.4% | estimated ± 6.2 pp, medium confidence |
| GPT-5.4 Pro | 83.3% | 55.9% | estimated ± 6.2 pp, medium confidence |
| GPT-5.5 Pro | 84.2% | 56.2% | estimated ± 6.2 pp, medium confidence |