Step-level classification on ProcessBench (test)
75.1F1 ScoreMCNIG
Evaluation results
| Method | Links | |
|---|---|---|
| MCNIGmodel size=14B2026.03 | 75.1 | |
| QwenPRMmodel size=7B2026.03 | 75 | |
| MCNIGmodel size=8B2026.03 | 73.2 | |
| IGmodel size=8B2026.03 | 70.1 | |
| Granite PRM v22026.03 | 51.9 | |
| MathShepherd2026.03 | 41.1 |