varies

SWE-bench is a benchmarking and evaluation platform for software engineering language models and agents. It provides datasets and curated leaderboards (including Verified, Multilingual, Lite, and Multimodal variants) to systematically compare the problem-solving capabilities and cost-efficiency of leading LMs and autonomous agents on real-world software engineering tasks. This tool is targeted at AI researchers, model developers, and organizations working on code generation and automation solutions.