SWE-bench
Indexed results from SWE-bench. Each ranking retains its own metric and evaluation conditions.
SWE-bench Verified
sweBench · SWE-bench · Resolved tasks · percent
- claude-opus-4-5@20251101anthropic74.4%
- gemini-3-pro-previewgoogle74.2%
- OpenAI: GPT-4oopenai71.2%
- OpenAI: GPT-5openai71.2%
- Anthropic: Claude Sonnet 4.5anthropic70.6%
- gemini-3-pro-previewgoogle69.6%
- OpenAI: GPT-5.2openai69%
- claude-haiku-4-5@20251001anthropic66.6%
- OpenAI: GPT-5.1openai66%
- MoonshotAI: Kimi K2 0711moonshotai65.4%
- OpenAI: GPT-5openai65%
- OpenAI: o3openai58.4%
Models
LoadingBenchmarks
Explore every benchmark
LoadingThese charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.