Shanghai Jiao Tong University and Team Unveil SWE-Explore Benchmark Revealing the Line-Level Localization Flaws of AI Coding Agents
An international team including Shanghai Jiao Tong University released SWE-Explore, a new benchmark that decouples code search and repair stages, first quantifying AI coding agents' weaknesses in line-level precision. It breaks the traditional single metric of final fix rate, offering a new standard for evaluating upstream search quality, advancing deeper AI software engineering assessment.....