Mathematical Reasoning on Olympiad standard (test)
38.6Pass@1AGPO
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AGPOBackbone=Qwen2.5-Math-7B2026.05 | 38.6 | 45.4 | 51.7 | 57.5 | 62 | 65.9 | 69.4 | 72.5 | 75.3 | |
| GRPOBackbone=Qwen2.5-Math-7B2026.05 | 37.5 | 44.2 | 50.3 | 55.7 | 60.2 | 63.8 | 66.9 | 69.5 | 72 | |
| PPOBackbone=Qwen2.5-Math-7B2026.05 | 37.1 | 43.7 | 49.8 | 55.1 | 59.6 | 63.2 | 66.3 | 69.1 | 71.9 | |
| W-REINFORCEBackbone=Qwen2.5-Math-7B2026.05 | 36.9 | 44.2 | 50.8 | 56.5 | 61.3 | 65.4 | 69 | 72.1 | 75.1 | |
| REINFORCEBackbone=Qwen2.5-Math-7B2026.05 | 35.1 | 42.5 | 47.9 | 52.3 | 57.8 | 61.4 | 66.7 | 69.3 | 72.5 | |
| Base ModelBackbone=Qwen2.5-Math-7B2026.05 | 25 | 35.2 | 44.4 | 52 | 58.3 | 63.4 | 67.6 | 71.3 | 74.4 |