Calibration
GPQA Diamond → AA-Omniscience Accuracy
AA-Omniscience Accuracy is estimated from GPQA Diamond with a inverse Michaelis–Menten curve fitted on 43 models measured on both: y = 0.05470·(x − 0.0000) / (0.0000 + 1.0417 − x), R² = 0.68, cross-validated error 8.6 pp. It is used for 2 estimates.
| Estimated model | GPQA Diamond | AA-Omniscience Accuracy | Source |
|---|---|---|---|
| LLaDA2.2-mini | 44.4% | 4.1% | estimated ± 8.6 pp, low confidence |
| Pareto 26.10 Preview | 92.4% | 42.9% | estimated ± 8.6 pp, medium confidence |