Skip to content

FrontierSWE v2

A 34-task expansion of FrontierSWE for ultra-long-horizon engineering and research work that remains far from saturation.

20 models

Higher is better.

  1. 1GPT-6 AstraOpenAI65.5
  2. 2Claude Opus 5.5Anthropic62.3
  3. 3Claude Sonnet 5.5Anthropic61.9
  4. 4Claude Fable 5.1Anthropic56.3
  5. 5Gemini 4 ArgonGoogle55.0
  6. 6Claude Opus 5Anthropic52.0
  7. 7Claude Fable 5Anthropic47.0
  8. 8Claude Haiku 5.5Anthropic43.8
  9. 9GPT-5.6 SolOpenAI32.2
  10. 10GLM-5.3Z.AI30.2
  11. 11Grok 4.7xAI29.5
  12. 12Kimi K3Moonshot AI25.9
  13. 13Muse Spark 1.3Meta25.8
  14. 14Grok 4.6xAI25.3
  15. 15Gemini 3.7 FlashGoogle20.3
  16. 16Gemini 3.8 FlashGoogle19.6
  17. 17GLM-5.3-FlashZ.AI18.1
  18. 18Qwen3.8 MaxAlibaba15.8
  19. 19Muse Spark 1.2Meta12.0
  20. 20InklingThinking Machines Lab4.1
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Five-trial mean task score (Mean@5), 0-100
Tasks
34 ultra-long-horizon engineering and research tasks
Authors
Proximal (2026)

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