ITG NowHome › AI glossary › SWE-bench
What is SWE-bench?
SWE-bench is a test that checks whether an AI can fix real bugs from real open source projects.
- Category Benchmarks
- Also called SWE-bench Verified
- Updated 20 Oct 2026
SWE-bench, explained simply
SWE-bench takes real problems from public code projects. The AI gets the code and a bug report and must write a fix.
The fix is checked by running the project's own tests. The score is the share of problems solved.
This matters because it looks like real coding work, not toy questions. Higher is better.
There are versions of different sizes, including a human checked one called Verified. Scores can differ between versions and setups.
On ITG Now, our daily AI news stories mention SWE-bench when a lab shares coding test results, with a link to the source.
Related terms
BenchmarkAI coding agentTerminal-BenchOpen sourceGitHubBenchmark contamination
Where you'll see it on ITG Now
- AI newsNew US lab gives away a giant AI free
- AI newsReflection Beam: open coding model arrives on AWS
Questions people ask about SWE-bench
Is a higher SWE-bench score better?
Yes. It means the AI fixed more of the real bugs.
What is SWE-bench Verified?
It is a smaller set that people checked by hand to make sure each problem is fair and solvable.
Does a high SWE-bench score mean the AI can replace a programmer?
No. It fixes small, clear bugs. Real jobs involve planning, talking to people and judgment.