Skip to content

JobBench

An occupational agent benchmark for professional workflows that workers say they most want delegated to AI.

34 models

Higher is better.

  1. 1Muse Wonder FullMeta75.3
  2. 2Muse WonderMeta70.5
  3. 3Muse Spark 1.3Meta64.9
  4. 4Ultramarine 1 FlashPaper Instruments62.9
  5. 5MiMo-V2.6-ProXiaomi62.0
  6. 6Muse Spark OpenMeta61.8
  7. 7Hy4 previewTencent61.7
  8. 8MiMo-V2.6-FlashXiaomi61.2
  9. 9Step 5 PreviewStepFun59.0
  10. 10Qwen3.8-Flash-NextAlibaba55.7
  11. 11Muse Spark 1.1Meta54.7
  12. 12Qwen3.8 MaxAlibaba53.4
  13. 13Kimi K3Moonshot AI52.9
  14. 14Atria Dawn PreviewShanghai Artificial Intelligence Laboratory50.3
  15. 15Claude Opus 4.7 (Adaptive)Anthropic45.9
  16. 16GPT-5.5OpenAI42.7
  17. 17GPT-5.4OpenAI38.9
  18. 18Claude Sonnet 4.6Anthropic36.9
  19. 19Claude Opus 4.6Anthropic36.7
  20. 20GPT-5.2OpenAI34.3
  21. 21GPT-5.3 CodexOpenAI33.7
  22. 22Qwen3.8-27BAlibaba33.4
  23. 23Claude Opus 4.5Anthropic32.3
  24. 24Claude Sonnet 4.5Anthropic27.7
  25. 25GPT-5.1-CodexOpenAI26.2
  26. 26GPT-5.2-CodexOpenAI26.0
  27. 27Claude 4.1 OpusAnthropic21.9
  28. 28Qwen3.5 PlusAlibaba18.5
  29. 29Claude 4 SonnetAnthropic18.4
  30. 30Claude Haiku 4.5Anthropic16.0
  31. 31Gemini 3 ProGoogle11.4
  32. 32Gemini 3 FlashGoogle11.4
  33. 33Kimi K2.5Moonshot AI8.7
  34. 34GPT-5 (high)OpenAI8.5
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Agentic workplace deliverables
Tasks
130 tasks across 35 occupations
Authors
Yuetai Li, Yichen Feng, Zhangchen Xu, Zixian Ma, Kaiyuan Zheng, Fengqing Jiang, Xinghua Sun, Rulin Shao, Zichen Chen, Yue Huang, Xinyang Han, Brian Lee, Kayla Xu, Shenglai Zeng, Hang Hua, Xiangliang Zhang, Basel Alomair, Ranjay Krishna, Luke Zettlemoyer, Pang Wei Koh, Bhaskar Ramasubramanian, Luyao Niu, Xiang Yue, Radha Poovendran (2026)

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