Benchmark fixes improve trust in software engineering AI tests
SWE-Bench Pro Verified: A Reliable Benchmark for Software Engineering Agents
Artificial IntelligenceSoftware Engineering
Summary
Software engineering AI tools are tested using a benchmark called SWE-Bench Pro, but this benchmark has some problems that make AI tools look better than they really are. The authors found that some tests leaked answers or had confusing questions, which made the AI perform artificially well. They created SWE-Bench Pro Verified, a cleaner version that fixes these problems and provides a fairer test. When using this improved benchmark, some AI models scored lower, showing past scores may have been too optimistic. This better benchmark helps people understand how well AI tools really can write software.
software engineering agentsbenchmarkreward hackingleakagetask qualityevaluationtask refinementsoftware testingAI performance
Authors
Pujun Zheng, Zixin Shang, Shufan Jiang, Wenhui Tian, Dongsheng Zhu, Zerun Ma, Dingbo Yuan, Qi Zhang
Abstract
SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-level tasks. However, our analysis work show that its evaluation is undermined by two sources of unreliability: \textbf{reward hacking}, enabled by leakage of gold solutions or hidden evaluation information, and \textbf{task quality issues}, including misleading problem statements and improperly scoped tests. These issues can inflate benchmark performance and obscure agents' true coding ability. We present \textbf{SWE-Bench Pro Verified}, a verified version of SWE-Bench Pro that addresses both problems. Our approach combines \textbf{anti-hacking} safeguards that eliminate major leakage channels without disrupting normal agent functionality, with \textbf{task refinement} that minimally corrects inconsistencies within flawed instances. Evaluations on SWE-Bench Pro Verified reveal that some models perform substantially worse than previously reported, suggesting that existing results on SWE-Bench Pro may overestimate real software engineering capability. SWE-Bench Pro Verified offers a more trustworthy benchmark for assessing software engineering agents.