Introducing SWE-bench Verified

In three linesOpenAI releases SWE-bench Verified, a human-validated subset of SWE-bench to more reliably evaluate AI models' ability to solve real-world software issues.

Summary generated by Claude — human-verified