Humanity's Last Exam
2,500 expert-written questions across 100+ subjects (Scale AI / CAIS).
- Published by
- Scale AI / CAISdata via Epoch AI Benchmarking Hub
- Category
- Reasoning
- Index weight
- 1.0
- Models
- 44
- Data as of
- 9 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
- 1Claude Fable 5.146.5%
- 2Gemini 3.1 Pro46.4%
- 3GPT-5.4 Pro44.3%
- 4Muse Spark40.6%
- 5Gemini 3 Pro37.5%
- 6GPT-5.436.2%
- 7Claude Opus 4.736.2%
- 8Claude Opus 4.634.4%
- 9GPT-5 Pro31.6%
- 10GPT-5.227.8%
- 11GPT-525.3%
- 12GPT-525.3%
- 13Claude Opus 4.525.2%
- 14Kimi K2.524.4%
- 15GPT-5.123.7%
44 of 44
| # | |||||
|---|---|---|---|---|---|
| 1 |  Claude Fable 5.1Anthropic | 46.5% | – | 70.3 | – |
| 2 |  Gemini 3.1 ProGoogle | 46.4% | – | 63.9 | – |
| 3 |  GPT-5.4 ProOpenAI | 44.3% | – | 64.2 | – |
| 4 |  Muse SparkMeta | 40.6% | – | 65.4 | – |
| 5 | Gemini 3 ProGoogle | 37.5% | – | 61.6 | – |
| 6 | GPT-5.4OpenAI | 36.2% | – | 61.3 | – |
| 7 | Claude Opus 4.7Anthropic | 36.2% | Claude Opus 4.7 | 65.5 | – |
| 8 | Claude Opus 4.6Anthropic | 34.4% | – | 54.1 | – |
| 9 | GPT-5 ProOpenAI | 31.6% | – | 60.7 | – |
| 10 | GPT-5.2OpenAI | 27.8% | – | 61.5 | – |
| 11 | GPT-5OpenAI | 25.3% | – | 59.5 | – |
| 12 | GPT-5OpenAI | 25.3% | – | 54.5 | – |
| 13 | Claude Opus 4.5Anthropic | 25.2% | – | 58.0 | – |
| 14 |  Kimi K2.5Moonshot AIopen | 24.4% | kimi-k2.5 | 56.1 | – |
| 15 | GPT-5.1OpenAI | 23.7% | – | 59.4 | – |
| 16 | Gemini 2.5 ProGoogle | 21.6% | – | 54.1 | – |
| 17 | o3OpenAI | 20.3% | – | 52.9 | – |
| 18 | GPT-5 miniOpenAI | 19.4% | – | 55.8 | – |
| 19 | o3OpenAI | 19.2% | – | 53.6 | – |
| 20 | Claude Opus 4.6Anthropic | 19.0% | – | 63.1 | – |
| 21 | o4-miniOpenAI | 18.1% | – | 48.0 | – |
| 22 | o4-miniOpenAI | 14.3% | – | 52.4 | – |
| 23 | Claude Sonnet 4.5Anthropic | 13.7% | – | 52.3 | – |
| 24 | Gemini 2.5 FlashGoogle | 12.1% | – | 48.8 | – |
| 25 | Claude Opus 4.1Anthropic | 11.5% | – | 52.6 | – |
| 26 | Claude Opus 4Anthropic | 10.7% | – | 50.7 | – |
| 27 | Gemini 3.1 Flash LiteGoogle | 8.6% | – | 54.3 | – |
| 28 |  GLM-4.5Zhipu AIopen | 8.3% | glm-4p5 | 50.2 | – |
| 29 | GLM-4.5-AirZhipu AIopen | 8.1% | – | 46.0 | – |
| 30 | o1-proOpenAI | 8.1% | – | – | – |
| 31 | Claude 3.7 SonnetAnthropic | 8.0% | – | 49.8 | – |
| 32 | o1OpenAI | 8.0% | – | 53.2 | – |
| 33 | Claude Sonnet 4Anthropic | 7.8% | – | 49.5 | – |
| 34 | GPT-5.1OpenAI | 6.8% | – | 46.6 | – |
| 35 | Gemini 2.0 FlashGoogle | 6.6% | – | 45.1 | – |
| 36 | Llama 4 MaverickMetaopen | 5.7% | Llama 4 Maverick | 42.6 | – |
| 37 | GPT-4.5OpenAI | 5.4% | – | 49.6 | – |
| 38 | GPT-4.1OpenAI | 5.4% | – | 47.3 | – |
| 39 | Gemini 1.5 Pro 002Google | 4.6% | – | 42.8 | – |
| 40 |  Mistral Medium (latest)Mistral AI | 4.5% | medium, 2505 | 45.8 | – |
| 41 |  Nova ProAmazon | 4.4% | – | 41.0 | – |
| 42 | Claude 3.5 SonnetAnthropic | 4.1% | Claude 3.5 Sonnet (October 2024) | 45.1 | – |
| 43 | Amazon Nova LiteAmazon | 3.6% | – | 39.4 | – |
| 44 | GPT-4oOpenAI | 2.7% | – | 42.3 | – |