Skip to content
Hbench

SWE-bench Verified

swebench-verified@1.0

500 real GitHub issues, each resolved against its project's own test suite. Scores this high on a public set may reflect contamination, so read them as harness differences on one model.

How SWE-bench Verified measures up to the dataset checklist

Boards

Every board here runs on an allowlisted network.

The boards on SWE-bench Verified
ModelHighest pass rateAttemptsHarnessesUpdated
Claude Opus 5.5
DeepSeek Harness93.4%91.2–95.4
1219 Oct 2026
1 board with no runs yet