Software Engineering Problem Solving on SWE-Bench Verified
25.8Rounds TakenMini SWE Agent + LaMR
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Mini SWE Agent + LaMRBackbone Model=Claude Opus 4.6, Pruning Technique=LaMR2026.05 | 25.8 | 380 | 76 | 0.4 | 0.453 | 3 | |
| Mini SWE Agent + SWE-PrunerBackbone Model=Claude Opus 4.6, Pruning Technique=SWE-Pruner2026.05 | 26.8 | 383 | 76.6 | 1 | 0.498 | 6.6 | |
| Mini SWE AgentBackbone Model=Claude Opus 4.6, Pruning Technique=None2026.05 | 30.9 | 378 | 75.6 | — | 0.467 | — | |
| Mini SWE Agent + LaMRBackbone Model=Claude Sonnet 4.5, Pruning Technique=LaMR2026.05 | 39.6 | 359 | 71.8 | 1.2 | 0.633 | 30.5 | |
| Mini SWE Agent + SWE-PrunerBackbone Model=Claude Sonnet 4.5, Pruning Technique=SWE-Pruner2026.05 | 41.7 | 360 | 72 | 1.4 | 0.701 | 23.1 | |
| Mini SWE AgentBackbone Model=Claude Sonnet 4.5, Pruning Technique=None2026.05 | 51 | 353 | 70.6 | — | 0.911 | — |