End-to-End Code Reproduction and Ablation on GEARS End-to-End
100Task Success RateAblateCell
Evaluation Results
| Method | Links | |
|---|---|---|
| AblateCellStage=End-to-End2026.04 | 100 | |
| Human ExpertsStage=End-to-End2026.04 | 61.4 | |
| RepoMasterStage=End-to-End2026.04 | 25 | |
| Mini-SWE-AgentStage=End-to-End2026.04 | 12.5 |