SWE-bench Verified

Resolves real GitHub issues end-to-end: read the repo, edit code, and pass the hidden tests — an agentic coding benchmark.

Code Tool use Reasoning Long context

Autonomous coding agents Bug fixing Coding & code review

Higher is better · % resolved

1 Claude Opus 5 74.5%

2 Claude Sonnet 5 70.3%

3 Gemini 2.5 Pro 63.8%

4 DeepSeek-V4 Flash 49.2%