Mathematical Reasoning on Minerva (ACC, mean@8)
66.78ACCBase model
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Base modelBackbone=Qwen3-8B-Base2026.04 | 66.78 | 66.78 | |
| GRPOBackbone=Qwen3-8B-Base2026.04 | 48.62 | 48.62 | |
| DAPOBackbone=Qwen3-8B-Base2026.04 | 48.02 | 48.02 | |
| CDEBackbone=Qwen3-8B-Base2026.04 | 47.75 | 47.75 | |
| DiPOBackbone=Qwen3-8B-Base2026.04 | 47.75 | 47.75 | |
| DAPO w/ ELBackbone=Qwen3-8B-Base2026.04 | 47.56 | 47.56 | |
| DiPOBackbone=Qwen3-4B-Base2026.04 | 44.76 | 44.76 | |
| DAPO w/ ELBackbone=Qwen3-4B-Base2026.04 | 44.53 | 44.53 | |
| GRPOBackbone=Qwen3-4B-Base2026.04 | 44.39 | 44.39 | |
| DAPOBackbone=Qwen3-4B-Base2026.04 | 44.34 | 44.34 | |
| CDEBackbone=Qwen3-4B-Base2026.04 | 43.11 | 43.11 | |
| DAPO w/ ELBackbone=Qwen2.5-7B2026.04 | 39.65 | 39.65 | |
| DiPOBackbone=Qwen2.5-7B2026.04 | 37.59 | 37.59 | |
| DAPOBackbone=Qwen2.5-7B2026.04 | 37.55 | 37.55 | |
| GRPOBackbone=Qwen2.5-7B2026.04 | 36.95 | 36.95 | |
| CDEBackbone=Qwen2.5-7B2026.04 | 35.94 | 35.94 | |
| Base modelBackbone=Qwen3-4B-Base2026.04 | 24.82 | 24.82 | |
| Base modelBackbone=Qwen2.5-7B2026.04 | 17.19 | 17.19 |