Reasoning benchmark

Humanity's Last Exam leaderboard

23 models have a published score. Claude Opus 5.5 (Anthropic) leads with 64.4, and the spread to the lowest score, Gemma 4 26B A4B at 8.7, is 55.7 points. The best open-weight result, Kimi K3.1 Pro at 51.1, trails the best closed model, Claude Opus 5.5, by 13.3. The highest-scoring model you can deploy on OpenLLM Buddy is Kimi K3, ranked 11 of 23 with 43.5. Sources report different settings for this test, such as test year, version, or reasoning effort. Each setting is listed in the source table, so compare close scores with care.

Best openllm buddy catalog

43.5

Kimi K3

6 models scored

Best open weights

51.1

Kimi K3.1 Pro

8 models scored

Best closed api

64.4

Claude Opus 5.5

9 models scored

Humanity's Last Exam scores

Exact-match accuracy. Scores are low across the board, so small gaps between models matter.

  • OpenLLM Buddy catalog
  • Open weights
  • Closed API
  1. 1Claude Opus 5.5Humanity's Last Exam, no tools; max effort64.4
  2. 2Claude Fable 5.1Humanity's Last Exam, no tools; max effort60.9
  3. 3Gemini 4 ArgonHumanity's Last Exam, no tools; AA independent run (High)57.1
  4. 4Claude Sonnet 5.5Humanity's Last Exam, no tools; max effort56.9
  5. 5GPT-6 AstraHumanity's Last Exam, no tools; AA independent run (Max)54.7
  6. 6GPT-6.1 SolHumanity's Last Exam, no tools; AA independent run (Max)52.9
  7. 7Kimi K3.1 ProHLE-Full (2,500 q), no tools, max effort (vendor-reported)51.1
  8. 8Gemini 3.8 FlashHumanity's Last Exam, no tools; AA independent run (High)47.8
  9. 9Step 5 Previewno tools, High effort (vendor-reported)46.5
  10. 10Claude Haiku 5.5Humanity's Last Exam, no tools; max effort45.9
  11. 11Kimi K3HLE-Full, no tools, max effort43.5
  12. 12MiniMax M3.1 Flashno tools, text-only subset (2,158 q)43.4
  13. 13Grok 4.7Humanity's Last Exam, no tools; AA independent run (Xhigh)43.1
  14. 14DeepSeek V4 ProV4-Pro-0813, no tools42.7
  15. 15GLM 5.3no tools, text-only subset (2,158 q); competitor comparison column42.3
  16. 16DeepSeek V4.1 FlashPass@1, no tools, max effort (text-only subset: 39.1)36.8
  17. 17Reflection Beam 1.0no tools, BF1636.0
  18. 18Qwen3.8 27BAlibaba Qwen30.8
  19. 19Nemotron 3 Ultrano tools26.7
  20. 20Qwen3.6 27BAlibaba Qwen24.0
  21. 21Gemma 4 31Bno tools19.5
  22. 22gpt-oss-120bhigh effort, no tools14.9
  23. 23Gemma 4 26B A4Bno tools8.7

What it measures

About 2,500 questions across mathematics, the sciences, and the humanities, contributed by subject experts and filtered so that leading models of the time failed them. Some questions include images.

How it is scored

Exact-match accuracy. Scores are low across the board, so small gaps between models matter.

Why it matters

Most older benchmarks are saturated; HLE still separates frontier models from each other.

Official benchmark page

More reasoning tests

Sources

ModelTestScoreSettingSource
Claude Opus 5.5Humanity's Last Exam64.4Humanity's Last Exam, no tools; max effortAnthropic Claude Opus 5.5 System Card (Table 8.1.A)
Claude Fable 5.1Humanity's Last Exam60.9Humanity's Last Exam, no tools; max effortAnthropic Claude Fable 5.1 / Mythos 5.1 System Card (Table 8.1.A)
Gemini 4 ArgonHumanity's Last Exam57.1Humanity's Last Exam, no tools; AA independent run (High)Artificial Analysis
Claude Sonnet 5.5Humanity's Last Exam56.9Humanity's Last Exam, no tools; max effortAnthropic Claude Sonnet 5.5 System Card (Table 8.1.A)
GPT-6 AstraHumanity's Last Exam54.7Humanity's Last Exam, no tools; AA independent run (Max)Artificial Analysis
GPT-6.1 SolHumanity's Last Exam52.9Humanity's Last Exam, no tools; AA independent run (Max)Artificial Analysis
Kimi K3.1 ProHumanity's Last Exam51.1HLE-Full (2,500 q), no tools, max effort (vendor-reported)DataCamp (relaying Moonshot's release table)
Gemini 3.8 FlashHumanity's Last Exam47.8Humanity's Last Exam, no tools; AA independent run (High)Artificial Analysis
Step 5 PreviewHumanity's Last Exam46.5no tools, High effort (vendor-reported)cellcog.ai (relaying StepFun's release table)
Claude Haiku 5.5Humanity's Last Exam45.9Humanity's Last Exam, no tools; max effortAnthropic Claude Haiku 5.5 System Card (Table 8.1.A); also in launch post
Kimi K3Humanity's Last Exam43.5HLE-Full, no tools, max effortKimi K3 model card (Hugging Face)
MiniMax M3.1 FlashHumanity's Last Exam43.4no tools, text-only subset (2,158 q)Hugging Face model card (MiniMax-M3.1-Flash)
Grok 4.7Humanity's Last Exam43.1Humanity's Last Exam, no tools; AA independent run (Xhigh)Artificial Analysis
DeepSeek V4 ProHumanity's Last Exam42.7V4-Pro-0813, no toolsHugging Face model card (DeepSeek-V4-Pro-0813)
GLM 5.3Humanity's Last Exam42.3no tools, text-only subset (2,158 q); competitor comparison columnMiniMax M3.1 Flash model card (comparison column)
DeepSeek V4.1 FlashHumanity's Last Exam36.8Pass@1, no tools, max effort (text-only subset: 39.1)DeepSeek-V4.1-Flash model card (Hugging Face)
Reflection Beam 1.0Humanity's Last Exam36.0no tools, BF16Hugging Face model card (Beam-1.0-501B-A23B)
Qwen3.8 27BHumanity's Last Exam30.8—Qwen3.8-27B model card (Hugging Face)
Nemotron 3 UltraHumanity's Last Exam26.7no toolsHugging Face model card (NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16)
Qwen3.6 27BHumanity's Last Exam24.0—Qwen3.6-27B model card (Hugging Face)
Gemma 4 31BHumanity's Last Exam19.5no toolsGemma 4 31B IT model card (Hugging Face)
gpt-oss-120bHumanity's Last Exam14.9high effort, no toolsgpt-oss model card paper (arXiv 2508.10925), Table 3
Gemma 4 26B A4BHumanity's Last Exam8.7no toolsGemma 4 31B IT model card (Hugging Face), 26B A4B column

Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works