Software Engineering on SWE-rebench 9B
38.16Avg@8 BC%EvoTrainer
Evaluation Results
| Method | Links | |
|---|---|---|
| EvoTrainerConfiguration=Proposed Method2026.06 | 38.16 | |
| RAGEN v2 SNR FilteringConfiguration=Algorithmic Baseline2026.06 | 35.74 | |
| Human-engineered RLConfiguration=Reference2026.06 | 33.77 | |
| AutoResearchConfiguration=Reference2026.06 | 33.33 | |
| RAGEN v1 FilteringConfiguration=Algorithmic Baseline2026.06 | 32.89 | |
| GRPO + Clip-HigherConfiguration=Algorithmic Baseline2026.06 | 32.79 | |
| Group-Variance FilteringConfiguration=Algorithmic Baseline2026.06 | 32.31 | |
| GSPO + Clip-HigherConfiguration=Algorithmic Baseline2026.06 | 31.97 | |
| Base model (no RL)RL=None2026.06 | 30.19 |