RepairBench

Leaderboard of frontier models for program repair.

If you use RepairBench, please cite:

@inproceedings{repairbench,
  title={RepairBench: Leaderboard of Frontier Models for Program Repair}, 
  author={André Silva and Martin Monperrus},
  booktitle = {IEEE/ACM International Workshop on Large Language Models for Code (LLM4Code)},
  year={2025},
  url={https://arxiv.org/abs/2409.18952}, 
  doi = {10.1109/LLM4Code66737.2025.00006}
}

For the code to reproduce the benchmark, please refer to https://github.com/ASSERT-KTH/repairbench-framework

Structure

• results includes all prompts, patches, and evaluation results
• scripts contains scripts used to parse results into other formats
• website contains the leaderboard's website code