# OpenAI Releases SWE-bench Verified to Better Test AI Coding Abilities

OpenAI announced the release of SWE-bench Verified, a refined version of the popular AI coding benchmark that promises more accurate evaluation of how well AI models can solve real-world software problems.

The key improvement is human validation. While the original SWE-bench dataset contained real GitHub issues, some entries had ambiguities or quality concerns that could skew results. The new verified subset has been carefully reviewed by humans to ensure each problem is clear, solvable, and truly representative of actual software engineering work.

This matters because AI coding assistants are rapidly improving, and developers need reliable benchmarks to compare different models. When evaluation data contains errors or unclear problems, it becomes difficult to know which AI systems genuinely perform better at helping with real coding tasks.

SWE-bench has become an industry-standard test for measuring AI coding capabilities