Calibration
GPQA Diamond → MMLU-Pro
MMLU-Pro is estimated from GPQA Diamond with a Michaelis–Menten curve fitted on 22 models measured on both: y = 2.0000·x / (1.16460 + x), R² = 0.92, cross-validated error 5.4 pp. It is used for 2 estimates.
| Estimated model | GPQA Diamond | MMLU-Pro | Source |
|---|---|---|---|
| LLaDA2.2-mini | 44.4% | 55.2% | estimated ± 5.4 pp, medium confidence |
| Pareto 26.10 Preview | 92.4% | 88.5% | estimated ± 5.4 pp, medium confidence |