SWE-bench
SWE-bench results reported for 98 models across 2 versions. Updated 2026-10-08.
Report an issueVersions
2 versions, plus 14 variants and configurations · Metric: Resolved Tasks
| Version | Released | Models | Results | ||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
SWE-bench Verified9 variants of Verified
| August 13, 2024 | 45 | 51 | ||||||||||||||||||||||||||||||||||||||||||||||||||
SWE-bench MultilingualLatest5 variants of Multilingual
| May 6, 2025 | 31 | 35 |
Recently reported results: SWE-bench Multilingual
Latest version: SWE-bench Multilingual