Skip to content

OSWorld 2.0

OSWorld 2.0 evaluates long-horizon computer-use tasks. Published results use different task releases, metrics, and run settings, so scores appear by comparable protocol and unmatched rows stay unranked.

24 models

Higher is better.

  1. 1GPT-6 AstraOpenAI72.6
  2. 2Claude Opus 5Anthropic70.6
  3. 3Gemini 4 ArgonGoogle69.2
  4. 4Muse Spark 1.3Meta66.9
  5. 5GPT-5.6 SolOpenAI62.6
  6. 6MiniMax M3.1 FlashMiniMax61.7
  7. 7GPT-6 SolOpenAI60.5
  8. 8Gemini 3.8 FlashGoogle59.0
  9. 9GPT-5.6 TerraOpenAI50.2
  10. 10Claude Opus 5.5Anthropic48.7
  11. 11Gemini 3.7 FlashGoogle47.9
  12. 12GPT-5.6 LunaOpenAI45.6
  13. 13Claude Fable 5.1Anthropic41.7
  14. 14Claude Opus 4.8Anthropic20.6
  15. 15Qwen3.8 MaxAlibaba19.4
  16. 16Qwen3.8-Flash-NextAlibaba19.4
  17. 17Claude Opus 4.7 (Adaptive)Anthropic18.2
  18. 18Muse Spark 1.1Meta14.2
  19. 19Claude Opus 4.7Anthropic13.9
  20. 20GPT-5.5OpenAI13.0
  21. 21Claude Sonnet 4.6Anthropic8.3
  22. 22Kimi K2.6Moonshot AI4.6
  23. 23MiniMax M3MiniMax4.6
  24. 24Qwen3.7 PlusAlibaba2.8
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Interactive computer-use evaluation; binary completion and partial-score metrics
Tasks
108 long-horizon computer-use workflows
Authors
Mengqi Yuan, Zilong Zhou, Xinzhuang Xiong, Weiming Wu, Jiayang Sun, Jiamin Song, Kaiqian Cui, Bowen Wang, Haoyuan Wu, Yitong Li, Dunjie Lu, Haikong Lu, Qi Zhen, Xinyuan Wang, Jiaqi Deng, Yuhao Yang, Cheng Chen, Boyuan Zheng, Alex Su, Xiao Yu, Hao Zou, Saaket Agashe, Xing Han Lu, Manpreet Kaur, Zhengyang Qi, Vincent Sunn Chen, Frederic Sala, Dayiheng Liu, Junyang Lin, Zhou Yu, Yu Su, Siva Reddy, Xin Eric Wang, Peng Qi, Tianbao Xie, Tao Yu (2026)

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