Mathematical Reasoning on GSM8K (Pass@1 Accuracy & Step Count)
86.5Pass@1 AccuracyHEX
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HEXGen. Len.=256, Training Necessity=Training-free2026.02 | 86.5 | 3,200 | |
| Majority VotingGen. Len.=256, Training Necessity=Training-free2026.02 | 84.69 | 640 | |
| DVOTINGGen. Len.=256, Training Necessity=Training-free2026.02 | 84.53 | 223.1 | |
| HEXGen. Len.=128, Training Necessity=Training-free2026.02 | 82.41 | 1,600 | |
| RFGGen. Len.=256, Training Necessity=FT or RL2026.02 | 82.1 | 512 | |
| DVOTINGGen. Len.=128, Training Necessity=Training-free2026.02 | 79.15 | 158.8 | |
| Majority VotingGen. Len.=128, Training Necessity=Training-free2026.02 | 78.47 | 320 | |
| LLaDA-1.5*Gen. Len.=256, Training Necessity=None2026.02 | 78.01 | 256 | |
| LLaDA-1.5*Gen. Len.=128, Training Necessity=None2026.02 | 72.48 | 128 |