Code Generation on LiveCodeBench v5 (held-out)
46.8Pass@1HSD
Evaluation Results
| Method | Links | |
|---|---|---|
| HSDBase Model=Qwen3-32B-Base2026.06 | 46.8 | |
| GRPO + OPSDBase Model=Qwen3-32B-Base2026.06 | 44.1 | |
| RLSDBase Model=Qwen3-32B-Base2026.06 | 43.9 | |
| SDPOBase Model=Qwen3-32B-Base2026.06 | 43.7 | |
| OPSDBase Model=Qwen3-32B-Base2026.06 | 43 | |
| DAPOBase Model=Qwen3-32B-Base2026.06 | 42.9 | |
| GSPOBase Model=Qwen3-32B-Base2026.06 | 42.6 | |
| Dr. GRPOBase Model=Qwen3-32B-Base2026.06 | 42.2 | |
| GRPOBase Model=Qwen3-32B-Base2026.06 | 41.6 | |
| HSDBase Model=Qwen3-8B-Base2026.06 | 37.2 | |
| GRPO + OPSDBase Model=Qwen3-8B-Base2026.06 | 35.1 | |
| RLSDBase Model=Qwen3-8B-Base2026.06 | 34.8 | |
| SDPOBase Model=Qwen3-8B-Base2026.06 | 34.5 | |
| OPSDBase Model=Qwen3-8B-Base2026.06 | 33.7 | |
| DAPOBase Model=Qwen3-8B-Base2026.06 | 33.4 | |
| GSPOBase Model=Qwen3-8B-Base2026.06 | 33 | |
| Dr. GRPOBase Model=Qwen3-8B-Base2026.06 | 32.6 | |
| GRPOBase Model=Qwen3-8B-Base2026.06 | 32 |