Calibration
GPQA Diamond → CritPt
CritPt is estimated from GPQA Diamond with a inverse Michaelis–Menten curve fitted on 44 models measured on both: y = 0.31779·(x − 0.8364) / (0.8364 + 0.2468 − x), R² = 0.79, cross-validated error 4.9 pp. It is used for 6 estimates.
| Estimated model | GPQA Diamond | CritPt | Source |
|---|---|---|---|
| Grok 4.20 | 88.5% | 7.8% | estimated ± 4.9 pp, high confidence |
| LLaDA2.2-mini | 44.4% | 0.0% | estimated ± 4.9 pp, medium confidence |
| MiniCPM5-1B | 26.3% | 0.0% | estimated ± 4.9 pp, medium confidence |
| Pareto 26.10 Preview | 92.4% | 17.5% | estimated ± 4.9 pp, high confidence |
| Beam | 90.5% | 12.2% | estimated ± 4.9 pp, high confidence |
| Solar Open 2 | 86.3% | 3.8% | estimated ± 4.9 pp, high confidence |