Skip to content

Humanity's Last Exam

An expert-authored benchmark designed to probe frontier knowledge and reasoning. BenchLM keeps protocol differences visible because tool-assisted and closed-book HLE runs answer different questions.

69 models

Higher is better.

  1. 1Claude Fable 5.1Anthropic65.0
  2. 2Claude Opus 5Anthropic64.7
  3. 3Claude Mythos 5Anthropic64.5
  4. 4Muse Spark 1.1Meta62.1
  5. 5GPT-5.4 ProOpenAI58.7
  6. 6Claude Opus 4.8Anthropic57.9
  7. 7Claude Sonnet 5Anthropic57.4
  8. 8GPT-5.5 ProOpenAI57.2
  9. 9Apodex 1.1Apodex56.1
  10. 10Kimi K3Moonshot AI56.0
  11. 11Hy4 previewTencent55.4
  12. 12Claude Opus 4.7 (Adaptive)Anthropic54.7
  13. 13Claude Opus 4.6Anthropic53.0
  14. 14dots3-note PreviewDots Studio52.6
  15. 15GPT-5.5OpenAI52.2
  16. 16GPT-5.4OpenAI52.1
  17. 17Muse SparkMeta50.4
  18. 18Claude Sonnet 4.6Anthropic49.0
  19. 19MiMo-V2.5-ProXiaomi48.0
  20. 20Inkling-SmallThinking Machines Lab47.8
  21. 21Agents-A1InternScience47.6
  22. 22Step 5 PreviewStepFun46.5
  23. 23InklingThinking Machines Lab46.0
  24. 24Ornith-1.5-397BOrnith AI44.6
  25. 25Qwen3.8 MaxAlibaba43.6
  26. 26DeepSeek V4 Pro 0813DeepSeek42.7
  27. 27DeepSeek V4.1 Flash 1017DeepSeek42.6
  28. 28GPT-5.4 miniOpenAI41.5
  29. 29Qwen3.7 MaxAlibaba41.4
  30. 30GLM-5.2Z.AI40.5
  31. 31Gemini 3.5 FlashGoogle40.2
  32. 32GPT-5.4 nanoOpenAI37.7
  33. 33Ling 3.1 FlashInclusionAI37.4
  34. 34Qwen3.8-Omni-FlashAlibaba36.5
  35. 35Qwen3.8-Flash-NextAlibaba35.9
  36. 36Grok 4.3xAI35.0
  37. 37DeepSeek V4 Flash 0731DeepSeek34.8
  38. 38Kimi K2.6Moonshot AI34.7
  39. 39Qwen3.7 PlusAlibaba34.7
  40. 40GLM-5.1Z.AI31.0
Higher is better. Showing the top 40 of 69. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Open-ended and multiple choice
Tasks
Expert-level questions
Authors
Center for AI Safety, Scale AI, and thousands of expert contributors (2025)

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

All results

1Claude Fable 5.1 Anthropic65.0
2Claude Opus 5 Anthropic64.7
3Claude Mythos 5 Anthropic64.5
4Muse Spark 1.1 Meta62.1
5GPT-5.4 Pro OpenAI58.7
6Claude Opus 4.8 Anthropic57.9
7Claude Sonnet 5 Anthropic57.4
8GPT-5.5 Pro OpenAI57.2
9Apodex 1.1 Apodex56.1
10Kimi K3 Moonshot AI56.0
11Hy4 preview Tencent55.4
12Claude Opus 4.7 (Adaptive) Anthropic54.7
13Claude Opus 4.6 Anthropic53.0
14dots3-note Preview Dots Studio52.6
15GPT-5.5 OpenAI52.2
16GPT-5.4 OpenAI52.1
17Muse Spark Meta50.4
18Claude Sonnet 4.6 Anthropic49.0
19MiMo-V2.5-Pro Xiaomi48.0
20Inkling-Small Thinking Machines Lab47.8
21Agents-A1 InternScience47.6
22Step 5 Preview StepFun46.5
23Inkling Thinking Machines Lab46.0
24Ornith-1.5-397B Ornith AI44.6
25Qwen3.8 Max Alibaba43.6
26DeepSeek V4 Pro 0813 DeepSeek42.7
27DeepSeek V4.1 Flash 1017 DeepSeek42.6
28GPT-5.4 mini OpenAI41.5
29Qwen3.7 Max Alibaba41.4
30GLM-5.2 Z.AI40.5
31Gemini 3.5 Flash Google40.2
32GPT-5.4 nano OpenAI37.7
33Ling 3.1 Flash InclusionAI37.4
34Qwen3.8-Omni-Flash Alibaba36.5
35Qwen3.8-Flash-Next Alibaba35.9
36Grok 4.3 xAI35.0
37DeepSeek V4 Flash 0731 DeepSeek34.8
38Kimi K2.6 Moonshot AI34.7
39Qwen3.7 Plus Alibaba34.7
40GLM-5.1 Z.AI31.0
41Qwen3.8-27B Alibaba30.8
42Claude Opus 4.5 Anthropic30.8
43GLM-5 Z.AI30.5
44Kimi K2.5 Moonshot AI30.1
45Nemotron 3.5 Super VL 120B A12B NVIDIA29.2
46Qwen3.6 Plus Alibaba28.8
47Qwen3.5 397B Alibaba28.7
48MiniMax M2.7 MiniMax28.0
49A.X K2 SK Telecom27.8
50Nemotron 3 Ultra NVIDIA26.7
51Gemma 4 31B Google26.5
52Ornith-1.5-35B-A3B Ornith AI25.6
53Hy3 Preview Tencent25.5
54GLM-4.7 Z.AI24.8
55Qwen3.6-27B Alibaba24.0
56Ling 3.0 Flash InclusionAI22.7
57Qwen3.6-35B-A3B Alibaba21.4
58Ornith-1.5-9B Ornith AI20.2
59Gemini 2.5 Pro Google18.8
60K-EXAONE 2.0 LG AI Research18.3
61Nemotron 3 Super 100B NVIDIA18.3
62Gemma 4 26B A4B Google17.2
63DeepSeek V3.1 (Reasoning) DeepSeek15.9
64GPT-OSS 120B OpenAI14.9
65GPT-OSS 20B OpenAI10.9
66Nemotron 3 Nano 30B NVIDIA10.6
67MiniCPM5-2B OpenBMB8.9
68LFM2.5-24B-A2B Liquid AI7.4
69Kimi K2 Moonshot AI4.7