Calibration
GPQA → AA-Omniscience Accuracy
AA-Omniscience Accuracy is estimated from GPQA with a offset logistic curve fitted on 60 models measured on both: y = 0.1622 + (1.2000 − 0.1622) / (1 + exp(−20.41·(x − 0.9741))), R² = 0.67, cross-validated error 8.1 pp. It is used for 15 estimates.
| Estimated model | GPQA | AA-Omniscience Accuracy | Source |
|---|---|---|---|
| Claude 3.5 Sonnet | 59.4% | 16.3% | estimated ± 8.1 pp, medium confidence |
| Claude Mythos 5 | 94.1% | 51.2% | estimated ± 8.1 pp, medium confidence |
| Interfaze Beta | 89.9% | 34.7% | estimated ± 8.1 pp, medium confidence |
| LFM2.5-230M | 25.4% | 16.2% | estimated ± 8.1 pp, low confidence |
| LFM2.5-VL-450M | 25.7% | 16.2% | estimated ± 8.1 pp, low confidence |
| MAI-Thinking-1 | 84.2% | 22.8% | estimated ± 8.1 pp, medium confidence |
| Ornith-1.5-35B-A3B | 89.2% | 32.6% | estimated ± 8.1 pp, medium confidence |
| Ornith-1.5-397B | 92.8% | 45.4% | estimated ± 8.1 pp, medium confidence |
| Ornith-1.5-9B | 86.4% | 26.1% | estimated ± 8.1 pp, medium confidence |
| Qwen3.8-Omni-Flash | 91.0% | 38.3% | estimated ± 8.1 pp, medium confidence |
| Sakana Fugu | 95.5% | 58.1% | estimated ± 8.1 pp, medium confidence |
| Sakana Fugu-Ultra | 95.5% | 58.1% | estimated ± 8.1 pp, medium confidence |
| Soofi S 30B-A3B | 43.4% | 16.2% | estimated ± 8.1 pp, medium confidence |
| ZAYA1-74B-Preview | 57.3% | 16.3% | estimated ± 8.1 pp, medium confidence |
| ZAYA1-8B | 71.0% | 16.7% | estimated ± 8.1 pp, medium confidence |