Mathematical Reasoning on Countdown-Stepwise (test)
80.37Pass@1REUSERL-SegCost
Evaluation results
| Method | Links | |
|---|---|---|
| REUSERL-SegCostBuffer=Global success buffer2026.05 | 80.37 | |
| Pure Round-Length2026.05 | 77.02 | |
| REUSERL-SegCost (no buffer)Buffer=None2026.05 | 68.94 | |
| Vanilla GRPO2026.05 | 68.46 | |
| Vanilla2026.05 | 20.57 |