Calibration
Vals GPQA Diamond → CritPt
CritPt is estimated from Vals GPQA Diamond with a inverse Michaelis–Menten curve fitted on 46 models measured on both: y = 0.13275·(x − 0.7437) / (0.7437 + 0.3212 − x), R² = 0.50, cross-validated error 7.2 pp. It is used for 8 estimates.
| Estimated model | Vals GPQA Diamond | CritPt | Source |
|---|---|---|---|
| GLM-5.3-Flash | 86.4% | 7.9% | estimated ± 7.2 pp, medium confidence |
| Qwen3.5 Flash | 82.8% | 4.7% | estimated ± 7.2 pp, medium confidence |
| MiMo-V2.5 | 81.6% | 3.9% | estimated ± 7.2 pp, medium confidence |
| Gemini 3.1 Flash-Lite | 81.1% | 3.5% | estimated ± 7.2 pp, medium confidence |
| Claude Haiku 4.5 | 72.2% | 0.0% | estimated ± 7.2 pp, medium confidence |
| GLM-4.5 | 72.2% | 0.0% | estimated ± 7.2 pp, medium confidence |
| Laguna M.1 | 27.0% | 0.0% | estimated ± 7.2 pp, low confidence |
| Laguna XS.2 | 55.1% | 0.0% | estimated ± 7.2 pp, medium confidence |