Skip to content

SWE-Rebench

A continuously updated software engineering benchmark by Nebius using fresh GitHub issues to avoid contamination. Models are evaluated 5 times per problem under a fixed ReAct scaffolding; the Resolved Rate (best pass@1) is reported.

13 models

Higher is better.

  1. 1Claude Opus 4.6Anthropic65.3
  2. 2GLM-5Z.AI62.8
  3. 3GLM-5.1Z.AI62.7
  4. 4DeepSeek V3.2DeepSeek60.9
  5. 5Claude Sonnet 4.6Anthropic60.7
  6. 6Qwen3.5-27BAlibaba58.9
  7. 7GLM-4.7Z.AI58.7
  8. 8Kimi K2.5Moonshot AI58.5
  9. 9GPT-5.3 CodexOpenAI58.2
  10. 10Composer 2Cursor58.0
  11. 11Qwen3.5-35B-A3BAlibaba53.7
  12. 12MiniMax M2.7MiniMax51.9
  13. 13Gemma 4 31BGoogle41.6
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Code patch generation
Tasks
Fresh GitHub issues (rolling window)
Authors
Nebius (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.