Process Reward Modeling Evaluation on SCIPRM-Bench
48.58ChemBench Tool-Calling F1Sci-PRM
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Sci-PRMModel Type=Open2026.06 | 48.58 | 53.62 | 84.57 | 56.19 | 83.33 | 76.91 | 27.14 | |
| Gemini-3-FlashModel Type=Closed2026.06 | 47.71 | 59.76 | 81.44 | 55.87 | 78.6 | 72.11 | 22.73 | |
| GLM-4.6VModel Type=Closed2026.06 | 47.11 | 42.75 | 59.75 | 46.89 | 85.4 | 73.07 | 38.51 | |
| GPT-5-MiniModel Type=Closed2026.06 | 46.3 | 49.87 | 81.56 | 53.38 | 82.94 | 75.92 | 29.56 | |
| Qwen3-VL-8BModel Type=Open2026.06 | 45.95 | 47.92 | 72.09 | 49.59 | 76.85 | 68.18 | 27.26 | |
| Doubao-Seed-1.6Model Type=Closed2026.06 | 43.9 | 50.92 | 80 | 51.29 | 76.21 | 68.51 | 24.92 | |
| Qwen3-VL-32BModel Type=Open2026.06 | 42.73 | 43.44 | 58.59 | 45.69 | 84.69 | 75.39 | 39 | |
| RandomModel Type=Baseline2026.06 | 38.45 | 33.43 | 32.42 | 35.23 | 60.83 | 53.27 | 25.6 | |
| LLaMA-3.2-11B-VModel Type=Open2026.06 | 18.68 | 36.44 | 46.24 | 32.13 | 57.54 | 52.81 | 25.41 |