Mathematical Reasoning on MATH (test) (AUROC, ECE, Recovery %)
66.8AUROCPost-solution probe
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Post-solution probeProbe Type=Post-solution2026.07 | 66.8 | 23.2 | — | |
| Distilled future confidence predictorProbe Type=Distilled2026.07 | 65.2 | 27.2 | — | |
| Pre-solution probeProbe Type=Pre-solution2026.07 | 63.4 | 30.6 | — |