Calibration
AA-GPQA Diamond → AA-HLE
AA-HLE is estimated from AA-GPQA Diamond with a offset logistic curve fitted on 179 models measured on both: y = 0.0435 + (0.8908 − 0.0435) / (1 + exp(−13.45·(x − 0.9351))), R² = 0.95, cross-validated error 4.0 pp. It is used for 2 estimates.
| Estimated model | AA-GPQA Diamond | AA-HLE | Source |
|---|---|---|---|
| o1-preview | 76.5% | 12.2% | estimated ± 4.0 pp, high confidence |
| o3-pro | 84.5% | 23.8% | estimated ± 4.0 pp, high confidence |