Calibration
Vals GPQA Diamond → CritPt
CritPt is estimated from Vals GPQA Diamond with a offset logistic curve fitted on 47 models measured on both: y = 0.0256 + (0.2754 − 0.0256) / (1 + exp(−33.91·(x − 0.9094))), R² = 0.50, cross-validated error 6.6 pp. It is used for 7 estimates.
| Estimated model | Vals GPQA Diamond | CritPt | Source |
|---|---|---|---|
| Claude Haiku 4.5 | 72.2% | 2.6% | estimated ± 6.6 pp, medium confidence |
| Gemini 3.1 Flash-Lite | 81.1% | 3.4% | estimated ± 6.6 pp, medium confidence |
| GLM-4.5 | 72.2% | 2.6% | estimated ± 6.6 pp, medium confidence |
| Laguna M.1 | 27.0% | 2.6% | estimated ± 6.6 pp, low confidence |
| Laguna XS.2 | 55.1% | 2.6% | estimated ± 6.6 pp, medium confidence |
| MiMo-V2.5 | 81.6% | 3.6% | estimated ± 6.6 pp, medium confidence |
| Qwen3.5 Flash | 82.8% | 4.0% | estimated ± 6.6 pp, medium confidence |