Language Understanding on MMLU 50/50 calibration (test)
73.7CR (%)Optimal pair
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Optimal pair2026.05 | 73.7 | 36 | |
| Optimal subsequence2026.05 | 73.7 | 36 | |
| Diagnostic learned router2026.05 | 73.7 | 39.3 | |
| Full fixed chain2026.05 | 59.3 | 25.9 | |
| Always-expensive2026.05 | 0 | — |