Reasoning on ProcessBench
69.85AccuracyQwen3-8B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-8BLoss function=base+aux2026.02 | 69.85 | — | |
| Qwen3-8BLoss function=base2026.02 | 67.8 | — | |
| Qwen3-4BLoss function=base+aux2026.02 | 67.42 | — | |
| Qwen3-4BLoss function=base2026.02 | 67.17 | — | |
| LLaMA3.1-8BLoss function=base+aux2026.02 | 31.14 | — | |
| SKBackbone=Llama3.1-8B-Instruct2025.12 | 29.49 | — | |
| PoEBackbone=Llama3.1-8B-Instruct2025.12 | 28.5 | — | |
| DIRBackbone=Llama3.1-8B-Instruct2025.12 | 27.73 | — | |
| InfoRMBackbone=Llama3.1-8B-Instruct2025.12 | 26.25 | — | |
| ALBMBackbone=Llama3.1-8B-Instruct2025.12 | 26.12 | — | |
| BaseBackbone=Llama3.1-8B-Instruct2025.12 | 25.39 | — | |
| LPBackbone=Llama3.1-8B-Instruct2025.12 | 24.91 | — | |
| LLaMA3.1-8BLoss function=base2026.02 | 23.63 | — | |
| DIRBackbone=OpenRLHF-Llama3-8B-SFT2025.12 | 13.82 | — | |
| ALBMBackbone=OpenRLHF-Llama3-8B-SFT2025.12 | 10.85 | — | |
| SKBackbone=OpenRLHF-Llama3-8B-SFT2025.12 | 10.31 | — | |
| PoEBackbone=OpenRLHF-Llama3-8B-SFT2025.12 | 9.68 | — | |
| LPBackbone=OpenRLHF-Llama3-8B-SFT2025.12 | 7.84 | — | |
| BaseBackbone=OpenRLHF-Llama3-8B-SFT2025.12 | 4.46 | — | |
| InfoRMBackbone=OpenRLHF-Llama3-8B-SFT2025.12 | 3.24 | — | |
| Gemini-3-FlashTier=High2026.06 | — | 90 | |
| GPT-5-miniTier=High2026.06 | — | 87.5 | |
| Seed2.0 Lite2026.06 | — | 92.4 | |
| Seed2.0 Mini2026.06 | — | 80.1 |