List your company

End-to-End Code Reproduction and Ablation on CPA End-to-End

74.8Task Success RateAblateCell
5.6423.59541.5559.505Apr 21, 2026

Evaluation results

MethodLinks
AblateCellStage=End-to-End2026.0474.8
Human ExpertsStage=End-to-End2026.0435.3
Mini-SWE-AgentStage=End-to-End2026.0416.7
RepoMasterStage=End-to-End2026.048.3