Can a model resolve real GitHub issues from popular Python repositories? Human-validated subset ensures accurate evaluation. Tests end-to-end software engineering ability.
Why it matters: The gold standard for real-world coding ability. Unlike HumanEval, tests understanding of large codebases, debugging, and complex changes. Scores range 20-80%.
Top Model
83.7%
Claude Opus 4.6
Average Score
65.3%
Across 22 models
Models Tested
22
Metric: resolve rate
Human Baseline
—
Range: 0%–100%
All models with a reported SWE-bench Verified score, ranked by highest resolve rate.
SWE-bench Verified is a standardized evaluation that measures AI model performance on specific tasks. It provides comparable scores across different models, helping developers choose the right model for their needs.
Claude Opus 4.6 currently holds the top score on the SWE-bench Verified benchmark. See our full rankings table above for the complete leaderboard with 22 models.
We update benchmark data from multiple sources including HuggingFace Open LLM Leaderboard and LMArena. Scores are refreshed regularly as new evaluations are published and new models are released.
No. While SWE-bench Verified is an important indicator, real-world performance depends on many factors including pricing, latency, context window, and specific task requirements. We recommend using our composite score which weighs multiple benchmarks and practical factors.