Mathematical Reasoning Process Evaluation on ProcessBench (test)
96.9GSM8K AccuracyQwen2.5-Math-7B-MathShepherd-r
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen2.5-Math-7B-MathShepherd-rModel Size=7B, Training Data=MathShepherd-r2026.04 | 96.9 | 51.7 | 67.4 | 20 | 96.8 | 33.2 | 7.3 | 95.9 | 13.5 | 4 | 96.7 | 7.6 | 30.4 | |
| Qwen2.5-Math-PRM-7BModel Size=7B2026.04 | 96.4 | 72 | 82.4 | 68 | 90.4 | 77.6 | 55.7 | 85.5 | 67.5 | 55.2 | 83 | 66.3 | 73.5 | |
| Qwen2.5-Math-PRM-7B-PDDL-rModel Size=7B, Training Data=PDDL-r2026.04 | 95.9 | 73.4 | 83.2 | 68.2 | 89.2 | 77.3 | 57.9 | 81.1 | 67.6 | 57.6 | 78.8 | 66.5 | 73.6 | |
| Qwen2.5-Math-7B-PRM800kModel Size=7B, Training Data=PRM800k2026.04 | 95.3 | 53.1 | 68.2 | 48 | 90.1 | 62.6 | 35.7 | 87.3 | 50.7 | 29.8 | 86.1 | 44.3 | 56.5 | |
| Qwen2.5-Math-7B-MathShepherd-PDDL-rModel Size=7B, Training Data=MathShepherd-PDDL-r2026.04 | 93.8 | 58 | 71.7 | 34.7 | 88.9 | 49.9 | 19.2 | 81.1 | 31.1 | 13.6 | 79.7 | 23.2 | 44 | |
| Llama-3.1-8B-PRM800k-PDDL-rModel Size=8B, Training Data=PRM800k-PDDL-r2026.04 | 91.7 | 58.5 | 71.4 | 53.9 | 55.4 | 54.6 | 46.3 | 27.1 | 34.2 | 42.7 | 32.4 | 36.8 | 49.3 | |
| Llama-3.1-8B-PRM800k-rModel Size=8B, Training Data=PRM800k-r2026.04 | 91.2 | 45.9 | 61.1 | 46 | 58.1 | 51.3 | 43.7 | 31.3 | 36.5 | 42.6 | 35.3 | 38.6 | 46.9 | |
| Qwen2.5-Math-7B-PRM800k-PDDL-rModel Size=7B, Training Data=PRM800k-PDDL-r2026.04 | 90.2 | 67.6 | 77.3 | 67.2 | 73.4 | 70.1 | 51.7 | 53.1 | 52.4 | 50.6 | 52.7 | 51.6 | 62.9 | |
| Qwen2.5-Math-7B-PRM800k-rModel Size=7B, Training Data=PRM800k-r2026.04 | 89.6 | 58 | 70.4 | 60.8 | 74.4 | 66.9 | 48.4 | 56 | 52 | 49.7 | 57.3 | 53.2 | 60.6 | |
| QwQ-32B-PreviewModel Category=Language Models as Critic2025.07 | 87.9 | — | 88 | — | 78.5 | 78.7 | — | 59.2 | 57.8 | — | 61.1 | 61.3 | 71.5 | |
| Llama-3.3-70B-InstructModel Category=Language Models as Critic2025.07 | 83.7 | — | 82.9 | — | 63.7 | 59.4 | — | 54.3 | 46.7 | — | 51 | 43 | 58 | |
| Skywork-PRM-7BModel Size=7B2026.04 | 82.9 | 61.8 | 70.8 | 43.8 | 62.2 | 53.6 | 17.9 | 31.9 | 22.9 | 14 | 41.9 | 21 | 42.1 | |
| GPT-4oModel Category=Language Models as Critic2025.07 | 80.2 | — | 79.2 | — | 63.4 | 63.6 | — | 50.1 | 51.4 | — | 50.1 | 53.5 | 61.9 | |
| Qwen2.5-72B-InstructModel Category=Language Models as Critic2025.07 | 77.9 | — | 76.2 | — | 65.4 | 61.8 | — | 59.8 | 54.6 | — | 55.1 | 52.2 | 61.2 | |
| Llama-3.1-70B-InstructModel Category=Language Models as Critic2025.07 | 75.3 | — | 74.9 | — | 52.6 | 48.2 | — | 50 | 46.7 | — | 43.2 | 41 | 52.7 | |
| Qwen2.5-Math-7B-PRM800KModel Category=Open-source PRM2025.07 | 73.5 | — | 68.2 | — | 65.1 | 62.6 | — | 53.2 | 50.7 | — | 43.4 | 44.3 | 56.5 | |
| Qwen2.5-14B-InstructModel Category=Language Models as Critic2025.07 | 72.3 | — | 69.3 | — | 59.2 | 53.3 | — | 50.2 | 45 | — | 43.5 | 41.3 | 52.2 | |
| CoLD-Base-PRM-7B(Ours)Model Category=Open-source PRM2025.07 | 72 | — | 68.6 | — | 67.3 | 67.7 | — | 54.6 | 56 | — | 47.8 | 51.3 | 60.9 | |
| Qwen2.5-Coder-32B-InstructModel Category=Language Models as Critic2025.07 | 72 | — | 68.9 | — | 64.5 | 60.1 | — | 57 | 48.9 | — | 52.5 | 46.3 | 56.1 | |
| Skywork-PRM-7BModel Category=Open-source PRM2025.07 | 71.6 | — | 70.8 | — | 54.5 | 53.6 | — | 25.6 | 22.9 | — | 23.7 | 21 | 42.1 | |
| Qwen2.5-32B-InstructModel Category=Language Models as Critic2025.07 | 70.6 | — | 65.6 | — | 61.9 | 53.1 | — | 53.5 | 40 | — | 47.7 | 38.3 | 49.3 | |
| Qwen2.5-Math-72B-InstructModel Category=Language Models as Critic2025.07 | 70.3 | — | 65.8 | — | 59.6 | 52.1 | — | 56.1 | 32.5 | — | 55.1 | 31.7 | 45.5 | |
| Qwen2-72B-InstructModel Category=Language Models as Critic2025.07 | 67.8 | — | 67.6 | — | 52.3 | 49.2 | — | 43.3 | 42.1 | — | 39.3 | 40.2 | 49.8 | |
| Meta-Llama-3-70B-InstructModel Category=Language Models as Critic2025.07 | 62.4 | — | 52.2 | — | 48.3 | 22.8 | — | 46.2 | 21.2 | — | 44.8 | 20 | 29.1 | |
| RLHFlow-PRM-Mistral-8BModel Category=Open-source PRM2025.07 | 62.3 | — | 50.4 | — | 42.1 | 33.4 | — | 22.3 | 13.8 | — | 19.1 | 15.8 | 28.4 | |
| Qwen2.5-Coder-14B-InstructModel Category=Language Models as Critic2025.07 | 61.9 | — | 50.1 | — | 54.2 | 39.9 | — | 51.4 | 34 | — | 55.6 | 27.3 | 37.8 | |
| Skywork-PRM-1.5BModel Category=Open-source PRM2025.07 | 59.9 | — | 59 | — | 49.1 | 48 | — | 20.5 | 19.3 | — | 19.7 | 19.2 | 36.4 | |
| Math-Shepherd-PRM-7BModel Category=Open-source PRM2025.07 | 58.3 | — | 47.9 | — | 45.1 | 29.5 | — | 39.7 | 24.8 | — | 34.8 | 23.8 | 31.5 | |
| RLHFlow-PRM-Deepseek-8BModel Category=Open-source PRM2025.07 | 56.9 | — | 38.8 | — | 45.1 | 33.8 | — | 26.5 | 16.9 | — | 23.2 | 16.9 | 26.6 | |
| Qwen2.5-Math-7B-InstructModel Category=Language Models as Critic2025.07 | 54.4 | — | 26.8 | — | 50.3 | 25.7 | — | 43.1 | 14.2 | — | 41.6 | 12.7 | 19.9 | |
| Qwen2.5-Coder-7B-InstructModel Category=Language Models as Critic2025.07 | 49.1 | — | 14.3 | — | 46.3 | 6.5 | — | 47.2 | 4.1 | — | 48.9 | 1.8 | 6.7 | |
| Qwen2.5-7B-InstructModel Category=Language Models as Critic2025.07 | 37.8 | — | 36.5 | — | 36.9 | 36.6 | — | 29.9 | 29.7 | — | 27.3 | 27.4 | 32.6 | |
| Llama-3.1-8B-InstructModel Category=Language Models as Critic2025.07 | 27.3 | — | 10.9 | — | 20.5 | 5.1 | — | 16 | 2.8 | — | 15 | 1.6 | 5.1 | |
| Meta-Llama-3-8B-InstructModel Category=Language Models as Critic2025.07 | 27.1 | — | 13.1 | — | 17.3 | 13.8 | — | 14.2 | 4.8 | — | 19.7 | 12.6 | 11.1 | |
| Qwen2-7B-InstructModel Category=Language Models as Critic2025.07 | 25.1 | — | 8.4 | — | 20.4 | 19 | — | 16.1 | 14.7 | — | 13.8 | 12.1 | 13.6 |