Step-wise Verification on ProcessBench OlympiadBench
61.3TNRCons@4
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Cons@4Backbone=Qwen3-8B2026.05 | 61.3 | 77.9 | 68.7 | |
| VerifySteer-BiBackbone=Qwen3-8B2026.05 | 60.1 | 76.7 | 67.4 | |
| VerifySteer-UniBackbone=Qwen3-8B2026.05 | 59.6 | 77 | 67.2 | |
| Qwen3-14BBackbone=Qwen3-14B2026.05 | 58.9 | 74.3 | 65.7 | |
| Prompt Opt.Backbone=Qwen3-8B2026.05 | 58.4 | 77.9 | 66.7 | |
| Basic PromptBackbone=Qwen3-8B2026.05 | 54 | 76.1 | 63.2 | |
| Qwen2.5-72BBackbone=Qwen2.5-72B2026.05 | 35.9 | 87.6 | 50.9 | |
| Llama-3.3-70BBackbone=Llama-3.3-70B2026.05 | 32.8 | 93.2 | 48.6 | |
| VerifySteer-UniBackbone=Qwen3-1.7B2026.05 | 15.6 | 56.6 | 24.4 | |
| VerifySteer-BiBackbone=Qwen3-1.7B2026.05 | 15 | 58.4 | 23.8 | |
| Prompt Opt.Backbone=Qwen3-1.7B2026.05 | 13.3 | 49.6 | 21 | |
| Cons@4Backbone=Qwen3-1.7B2026.05 | 11 | 58.5 | 18.5 | |
| Basic PromptBackbone=Qwen3-1.7B2026.05 | 7.7 | 78.8 | 14.1 |