Mathematical Reasoning on Olympiad Bench (accuracy, Average)
44AccuracyBRIDGE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BRIDGEBackbone=Qwen3-8B-Base2025.09 | 44 | 49.9 | |
| LUFFYBackbone=Qwen3-8B-Base2025.09 | 43.1 | 44 | |
| RLBackbone=Qwen3-8B-Base2025.09 | 42.4 | 42.9 | |
| CHORDBackbone=Qwen3-8B-Base2025.09 | 42.2 | 45.9 | |
| SRFTBackbone=Qwen3-8B-Base2025.09 | 40 | 39.8 | |
| SFT→RLBackbone=Qwen3-8B-Base2025.09 | 39.6 | 45.5 | |
| SFT+RLBackbone=Qwen3-8B-Base2025.09 | 39.2 | 40.1 | |
| SFTBackbone=Qwen3-8B-Base2025.09 | 29.8 | 37.6 | |
| BaseBackbone=Qwen3-8B-Base2025.09 | 22.5 | 26.6 |