Mathematical Reasoning on MATH500 (ACC, mean@8)
89.58Accuracy (MATH500)DAPO w/ EL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DAPO w/ ELBackbone=Qwen3-8B-Base2026.04 | 89.58 | 89.58 | |
| DiPOBackbone=Qwen3-8B-Base2026.04 | 89.55 | 89.55 | |
| DAPOBackbone=Qwen3-8B-Base2026.04 | 89.43 | 89.43 | |
| CDEBackbone=Qwen3-8B-Base2026.04 | 89.35 | 89.35 | |
| GRPOBackbone=Qwen3-8B-Base2026.04 | 89.08 | 89.08 | |
| DiPOBackbone=Qwen3-4B-Base2026.04 | 87 | 87 | |
| DAPO w/ ELBackbone=Qwen3-4B-Base2026.04 | 86.78 | 86.78 | |
| DAPOBackbone=Qwen3-4B-Base2026.04 | 86.43 | 86.43 | |
| CDEBackbone=Qwen3-4B-Base2026.04 | 85.93 | 85.93 | |
| GRPOBackbone=Qwen3-4B-Base2026.04 | 85.83 | 85.83 | |
| DiPOBackbone=Qwen2.5-7B2026.04 | 80.35 | 80.35 | |
| DAPO w/ ELBackbone=Qwen2.5-7B2026.04 | 79.85 | 79.85 | |
| GRPOBackbone=Qwen2.5-7B2026.04 | 79.15 | 79.15 | |
| DAPOBackbone=Qwen2.5-7B2026.04 | 79.08 | 79.08 | |
| CDEBackbone=Qwen2.5-7B2026.04 | 79 | 79 | |
| Base modelBackbone=Qwen3-8B-Base2026.04 | 66.78 | 66.78 | |
| Base modelBackbone=Qwen3-4B-Base2026.04 | 48.48 | 48.48 | |
| Base modelBackbone=Qwen2.5-7B2026.04 | 41.53 | 41.53 |