RepairBench: Leaderboard of Frontier Models for Program Repair
AI-driven program repair uses AI models to repair buggy software by producing patches. Rapid advancements in frontier models surely impact performance on the program repair task. Yet, there is a lack of frequent and standardized evaluations to actually understand the strengths and weaknesses of models. To that end, we propose RepairBench, a novel leaderboard for AI-driven program repair. The key characteristics of RepairBench are: 1) it is execution-based: all patches are compiled and executed against a test suite, 2) it assesses frontier models in a frequent and standardized way. RepairBench leverages two high-quality benchmarks, Defects4J and GitBug-Java, to evaluate frontier models only against real-world program repair tasks. At the time of writing, RepairBench shows that \textit{claude-3-5-sonnet-20241022} is the best model for program repair, and \textit{qwen-2.5-coder-32b-instruct} the cheapest while maintaining good performance. We publicly release the evaluation framework of RepairBench as well as all patches generated in the course of the evaluation.
Sat 3 MayDisplayed time zone: Eastern Time (US & Canada) change
09:00 - 10:30 | |||
09:00 10mDay opening | Opening LLM4Code Lingming Zhang University of Illinois at Urbana-Champaign, Prem Devanbu University of California at Davis, Zijian Wang AWS AI Labs | ||
09:10 60mKeynote | Keynote 1: Building the Hybrid Human-AI Developer: From Code Completion to Agents (zoom talk) LLM4Code Federico Cassano Cursor AI | ||
10:10 10mTalk | Are Large Language Models Memorizing Bug Benchmarks? LLM4Code Daniel Ramos Carnegie Mellon University, Claudia Mamede Carnegie Mellon University, Kush Jain Carnegie Mellon University, Paulo Canelas Carnegie Mellon University, Catarina Gamboa Carnegie Mellon University, Claire Le Goues Carnegie Mellon University | ||
10:20 10mTalk | RepairBench: Leaderboard of Frontier Models for Program Repair LLM4Code |