CigaR: Cost-efficient Program Repair with LLMs

Large language models (LLM) have proven to be effective at automated program repair (APR). However, using LLMs can be highly costly, with companies invoicing users by the number of tokens. In this paper, we propose CigaR, the first LLM-based APR tool that focuses on minimizing the repair cost. CigaR works in two major steps: generating a plausible patch and multiplying plausible patches. CigaR optimizes the prompts and the prompt setting to maximize the information given to LLMs in the smallest possible number of tokens. Our experiments on 267 bugs from the widely used Defects4J dataset shows that CigaR reduces the token cost by 62. On average, CigaR spends 171k tokens per bug while the baseline uses 451k tokens. On the subset of bugs that are fixed by both, CigaR spends 20k per bug while the baseline uses 695k tokens, a cost saving of 97. Our extensive experiments show that CigaR is a cost-effective LLM-based program repair tool that uses a low number of tokens to generate automatic patches.