Software Engineering Issue Solving on SWE-bench Verified (Pass@K)
51.2Pass@1HHD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HHDBackbone Model=Qwen-72b2026.05 | 51.2 | 67 | 70.2 | |
| HHD-MBackbone Model=Qwen-72b2026.05 | 47.2 | 61.8 | 67.8 | |
| Dense Expert JudgeBackbone Model=Qwen-72b2026.05 | 45.8 | 65 | 68.8 | |
| Agent-RLVR-PlanBackbone Model=Qwen-72b2026.05 | 45.6 | 65.4 | 70.2 | |
| Naive RFTBackbone Model=Qwen-72b2026.05 | 44 | 62.2 | 67.2 | |
| SE-agent-ReflectBackbone Model=Qwen-72b2026.05 | 43.4 | 64.4 | 69 | |
| Agent-RLVR-EFBackbone Model=Qwen-72b2026.05 | 43.4 | 65 | 69 | |
| HHDBackbone Model=GLM-Air2026.05 | 42 | 55.4 | 60.8 | |
| HHD-MBackbone Model=GLM-Air2026.05 | 41 | 54.6 | 59.6 | |
| SE-agent-ReflectBackbone Model=GLM-Air2026.05 | 40 | 55.2 | 59.8 | |
| Agent-RLVR-EFBackbone Model=GLM-Air2026.05 | 39.4 | 54.4 | 59.6 | |
| Dense Expert JudgeBackbone Model=GLM-Air2026.05 | 39.4 | 55.2 | 59 | |
| Agent-RLVR-PlanBackbone Model=GLM-Air2026.05 | 38.6 | 55 | 60.2 | |
| Naive RFTBackbone Model=GLM-Air2026.05 | 38 | 53 | 58.8 |