HLE (Humanity's Last Exam)Reasoning

Humanity's Last Exam (HLE) is a highly difficult benchmark released by CAIS and Scale AI in 2025. It contains about 2,500 challenging questions across mathematics, physics, chemistry, biology, history, law, and other fields, using multiple-choice and free-response formats to evaluate expert-level knowledge.

Ranking Sources

#ModelScore

1

Claude Fable 5Anthropic

53.3%

2

Claude Opus 4.8Anthropic

49.8%

3

GPT-5.6 SolOpenAI

47.2%

4

GPT-5.5OpenAI

46.9%

5

Claude Opus 4.7Anthropic

46.9%

6

Gemini 3.1 ProGoogle

45.0%

7

Qwen3.8-2.4T-A95BAlibaba

43.6%

8

Qwen3.8-MaxAlibaba

43.6%

9

Kimi K3Moonshot AI

43.5%

10

GPT-5.5 ProOpenAI

43.1%

69 models total

Back to benchmarks