Type: url
Original: https://www.swebench.com/
Status: generated
Summary
The home page of SWE-bench, hosting the official leaderboards for the SWE-bench family of benchmarks (the full benchmark of 2,294 real GitHub issues from 12 Python repositories, plus Verified, Bash Only, Lite, Multilingual and Multimodal subsets and variants), each ranked by the percentage of task instances resolved. It also lists news from the SWE-bench team — releases such as SWE-bench Lite, the Docker-ized harness, SWE-bench Verified with OpenAI, SWE-bench Multimodal, SWE-agent, mini-SWE-agent, SWE-smith, CodeClash and ProgramBench — and links related tools in the same family.