Skip to content

Humanity's Last Exam with tools

Tool-augmented Humanity's Last Exam scores reported in DeepSeek-V4 thinking-mode evaluations.

27 models

Higher is better.

  1. 1Claude Opus 5.5Anthropic67.7
  2. 2Claude Opus 5Anthropic64.7
  3. 3Claude Sonnet 5.5Anthropic64.5
  4. 4DeepSeek V4.1 Flash 1017DeepSeek63.9
  5. 5GLM-5.3Z.AI62.5
  6. 6Muse Spark OpenMeta61.0
  7. 7DeepSeek V4 Pro 0813DeepSeek60.0
  8. 8Claude Haiku 5.5Anthropic57.4
  9. 9Claude Sonnet 5Anthropic57.4
  10. 10GPT-6 AstraOpenAI57.2
  11. 11Qwen3.8 MaxAlibaba56.2
  12. 12Ornith-1.5-397BOrnith AI56.1
  13. 13Apodex 1.1Apodex56.1
  14. 14Hy4 previewTencent55.4
  15. 15GLM-5.3-FlashZ.AI55.3
  16. 16GLM-5.2Z.AI54.7
  17. 17Qwen3.7 MaxAlibaba53.5
  18. 18dots3-note PreviewDots Studio52.6
  19. 19GLM-5.1Z.AI52.3
  20. 20Beam 1.0 501B-A23BReflection AI51.1
  21. 21GLM-5Z.AI50.4
  22. 22Agents-A1InternScience47.6
  23. 23Step 3.7 FlashStepFun47.2
  24. 24DeepSeek V4 Flash 0731DeepSeek45.1
  25. 25Nemotron 3 UltraNVIDIA37.4
  26. 26Ornith-1.5-35B-A3BOrnith AI33.4
  27. 27Ornith-1.5-9BOrnith AI30.5
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Pass@1
Tasks
Expert questions with tool use
Authors
DeepSeek-AI (2026)

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