Mathematical Problem Solving on MATH 500 (test)
85.4Pass@1 AccPRM-CoT (Process-Aware)
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PRM-CoT (Process-Aware)Base Policy=Qwen2.5-Math-7B2025.12 | 85.4 | 83.6 | 91 | 92.2 | — | — | — | — | — | — | — | |
| RLVR (Ground Truth)Base Policy=Qwen2.5-Math-7B2025.12 | 83.7 | 82.5 | 91 | 92 | — | — | — | — | — | — | — | |
| PRM (Process-Aware)Base Policy=Qwen2.5-Math-7B2025.12 | 82.8 | 81.43 | 90.8 | 91.4 | — | — | — | — | — | — | — | |
| SFT (Baseline)Base Policy=Qwen2.5-Math-7B2025.12 | 79 | 78.2 | 84.8 | 88.6 | — | — | — | — | — | — | — | |
| GRPOBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.06 | 77.5 | — | 86.6 | 88.3 | 81.7 | 84.6 | 89.7 | 91 | 92 | 93 | 87.2 | |
| BBGBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.06 | 75.6 | — | 89.4 | 91.6 | 81.6 | 86.3 | 93.4 | 94.8 | 95.8 | 96.6 | 89.5 | |
| REINFORCEBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.06 | 75.5 | — | 85.9 | 87.9 | 79.7 | 82.7 | 88.8 | 89.5 | 91 | 92 | 85.9 | |
| W-RFBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.06 | 75.5 | — | 87 | 88.9 | 80.8 | 84.5 | 90.7 | 92.2 | 93.2 | 93.8 | 87.4 | |
| Base ModelBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.06 | 64.5 | — | 88.2 | 90.8 | 76.2 | 83.8 | 92.6 | 94 | 95.3 | 96.2 | 86.8 |