Mathematical Reasoning on AMC, AIME, HMMT, and OlympiadBench Aggregate
54.11AccuracyDAPO w/ EDAS
Evaluation Results
| Method | Links | |
|---|---|---|
| DAPO w/ EDASBackbone=Qwen3-8B2026.05 | 54.11 | |
| DAPO w/ EDASBackbone=Qwen3-4B2026.05 | 53.73 | |
| DAPOBackbone=Qwen3-4B2026.05 | 51.84 | |
| DAPOBackbone=Qwen3-8B2026.05 | 47.82 | |
| DAPO w/ En AdvBackbone=Qwen3-8B2026.05 | 44.78 | |
| DAPO w/ En AdvBackbone=Qwen3-4B2026.05 | 41.87 | |
| GRPO w/ EDASBackbone=Qwen3-8B2026.05 | 38.35 | |
| DAPO w/ EDASBackbone=Qwen3-4B-Base2026.05 | 37.02 | |
| GRPOBackbone=Qwen3-8B2026.05 | 36.75 | |
| PKPOBackbone=Qwen3-8B2026.05 | 36.57 | |
| BaselineBackbone=Qwen3-4B2026.05 | 33.68 | |
| BaselineBackbone=Qwen3-8B2026.05 | 33.63 | |
| DAPOBackbone=Qwen3-4B-Base2026.05 | 27.14 | |
| DAPO w/ En AdvBackbone=Qwen3-4B-Base2026.05 | 26.9 | |
| BaselineBackbone=Qwen3-4B-Base2026.05 | 12.93 |