Mathematical Reasoning on AIME (Best-of-64 accuracy)
50.98Best-of-64 AIME AccuracyMSV64
Evaluation Results
| Method | Links | |
|---|---|---|
| MSV64Best-of-K=642026.03 | 50.98 | |
| PRM-7BBest-of-K=64, Backbone=Qwen2.5-Math-7B2026.03 | 43.97 | |
| PRM-7B + WVBest-of-K=64, Backbone=Qwen2.5-Math-7B, Weighted Voting=true2026.03 | 41.74 | |
| QwenPRMModel Size=7B2026.03 | 37 | |
| IGModel Size=8B2026.03 | 35 | |
| MCNIGModel Size=8B2026.03 | 32.5 | |
| Majority voting2026.03 | 29 | |
| ORMModel Size=8B2026.03 | 29 | |
| Granite PRM v22026.03 | 27 | |
| OVM2026.03 | 26.1 | |
| MathShepherd2026.03 | 23 | |
| ImplicitPRM2026.03 | 19.6 | |
| Single sampling2026.03 | 17.4 |