SWE-Bench verified
SWE-Bench verified is a coding benchmark tracked on BenchGecko. Real-world software engineering tasks from GitHub issues. Models must diagnose bugs and write patches that pass test suites. Human-verified subset of SWE-bench.
Built from BenchGecko data as of October 2, 2026 · updates when the data changes
SWE-Bench verified is a coding benchmark tracked on BenchGecko. Real-world software engineering tasks from GitHub issues. Models must diagnose bugs and write patches that pass test suites. Human-verified subset of SWE-bench.
Basic
Real-world software engineering tasks from GitHub issues. Models must diagnose bugs and write patches that pass test suites. Human-verified subset of SWE-bench.
Deep
Scores are reported (%, maximum 100); higher is better. BenchGecko collects SWE-Bench verified scores from Epoch AI and refreshes them daily; the leaderboard on this page shows the current top models.
Expert
Compare models only on the same benchmark version and settings. A benchmark separates models well while scores are spread out; when top models bunch together near the maximum, it stops telling them apart.
Depending on why you're here
- ·coding benchmark (%, maximum 100)
- ·Source: Epoch AI
- ·Useful if your workload is coding
- ·Check the live leaderboard on /benchmark/swe-bench-verified before choosing a model
- ·Labs cite benchmark results in launch announcements; check the source and version
- ·SWE-Bench verified is a test of how good an AI is at coding
- ·A higher score means better results on that kind of task