Mathematical Reasoning on MATH levels 3–5 50/50 calibration (test)
81.2CR (%)Diagnostic learned router
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Diagnostic learned router2026.05 | 81.2 | 39.4 | |
| Optimal pair2026.05 | 79.5 | 39.3 | |
| Optimal subsequence2026.05 | 79.5 | 39 | |
| Full fixed chain2026.05 | 66.7 | 34.4 | |
| Always-expensive2026.05 | 0 | — |