Code Generation on HumanEval+ latest (held-out)
87.4Pass@1SDPO
Evaluation Results
| Method | Links | |
|---|---|---|
| SDPOBase Model=Qwen3-32B-Base2026.06 | 87.4 | |
| HSDBase Model=Qwen3-32B-Base2026.06 | 87.3 | |
| GRPO + OPSDBase Model=Qwen3-32B-Base2026.06 | 87.1 | |
| RLSDBase Model=Qwen3-32B-Base2026.06 | 87 | |
| OPSDBase Model=Qwen3-32B-Base2026.06 | 86.9 | |
| DAPOBase Model=Qwen3-32B-Base2026.06 | 86.8 | |
| GSPOBase Model=Qwen3-32B-Base2026.06 | 86.6 | |
| Dr. GRPOBase Model=Qwen3-32B-Base2026.06 | 86.5 | |
| GRPOBase Model=Qwen3-32B-Base2026.06 | 86.3 | |
| SDPOBase Model=Qwen3-8B-Base2026.06 | 82 | |
| HSDBase Model=Qwen3-8B-Base2026.06 | 81.9 | |
| GRPO + OPSDBase Model=Qwen3-8B-Base2026.06 | 81.7 | |
| RLSDBase Model=Qwen3-8B-Base2026.06 | 81.6 | |
| OPSDBase Model=Qwen3-8B-Base2026.06 | 81.4 | |
| DAPOBase Model=Qwen3-8B-Base2026.06 | 81.1 | |
| GSPOBase Model=Qwen3-8B-Base2026.06 | 80.9 | |
| Dr. GRPOBase Model=Qwen3-8B-Base2026.06 | 80.7 | |
| GRPOBase Model=Qwen3-8B-Base2026.06 | 80.5 |