Software Engineering Task Resolution on SWE-bench In-Distribution
52.34Resolved RateSkillFlow
Evaluation Results
| Method | Links | |
|---|---|---|
| SkillFlowMethod Category=Ours2026.05 | 52.34 | |
| FlowSteerMethod Category=Agent+RL2026.05 | 43.75 | |
| AgentFlowMethod Category=Agent+RL2026.05 | 41.41 | |
| SkillRLMethod Category=Agent+RL2026.05 | 39.06 | |
| v4-flashMethod Category=Baseline2026.05 | 38.28 | |
| Qwen3.5-AFlowMethod Category=AFlow, Backbone=Qwen3.52026.05 | 30.47 | |
| Qwen3.5-GRPOMethod Category=GRPO, Backbone=Qwen3.52026.05 | 18.75 | |
| Qwen3.5-9BMethod Category=Baseline2026.05 | 17.19 | |
| Qwen3.5-SFTMethod Category=SFT, Backbone=Qwen3.52026.05 | 16.41 |