わたしたちの信念

HLE (Humanity's Last Exam)

このベンチマークについて

人類の最高水準の専門知識を測る最高難度ベンチマーク。50以上の分野の極めて難しい問題で構成されます。スコアは正答率(%)です。

出典: Artificial Analysis
順位モデル
#1

Google

Gemini 3.1 Pro

44.7%
#2

OpenAI

GPT-5.4

41.6%
#3

Meta

Muse Spark

39.9%
#4

Anthropic

Claude Opus 4.7

39.6%
#5

Anthropic

Claude Opus 4.6

36.7%
#6

DeepSeek

DeepSeek V4 Pro

35.9%
#7

Google

Gemini 3 Flash

34.7%
#8

Grok

Grok 4.20 (Reasoning)

32.2%
#9

DeepSeek

DeepSeek V4 Flash

32.1%
#10

Moonshot AI

Kimi K2.5

29.4%
#11

Anthropic

Claude Opus 4.5

28.4%
#12

Xiaomi

MiMo-V2-Pro

28.3%
#13

MiniMax

MiniMax M2.7

28.1%
#14

Z.ai

GLM-5.1

28.0%
#15

Z.ai

GLM-5

27.2%
#16

Alibaba

Qwen3.6 Plus

25.7%
#17

Grok

Grok 4.20

24.2%
#18

Google

Gemma 4 31B

22.7%
#19

DeepSeek

DeepSeek V3.2

22.2%
#20

Google

Gemini 2.5 Pro

21.1%
#21

NVIDIA

Nemotron 3 Super

19.2%
#22

MiniMax

MiniMax M2.5

19.1%
#23

Alibaba

Qwen3.5 397B A17B

18.8%
#24

Grok

Grok 4.1 Fast (Reasoning)

17.6%
#25

Anthropic

Claude Sonnet 4.5

17.3%
#26

OpenAI

GPT-5.4 Mini

17.1%
#27

Google

Gemini 3.1 Flash Lite

16.2%
#28

OpenAI

GPT-5.4 Nano

14.7%
#29

Arcee AI

Trinity Large Thinking

14.7%
#30

Baidu

ERNIE 5.0 Thinking

12.7%
#31

Anthropic

Claude Opus 4.1

11.9%
#32

Anthropic

Claude Opus 4

11.7%
#33

Google

Gemini 2.5 Flash

11.1%
#34

Anthropic

Claude Sonnet 4.6

10.8%
#35

Anthropic

Claude Haiku 4.5

9.7%
#36

Anthropic

Claude Sonnet 4

9.6%
#37

Mistral AI

Mistral Small 4

9.5%
#38

OpenAI

GPT-5 Nano

7.6%
#39

Meituan

Longcat Flash Chat

6.0%
#40

OpenAI

GPT-5

5.8%
#41

OpenAI

GPT OSS 120B

5.2%
#42

OpenAI

GPT-5 Mini

5.0%
#43

Grok

Grok 4.1 Fast

5.0%
#44

Meta

Llama 4 Maverick

4.8%
#45

Google

Gemini 2.5 Flash Lite

4.6%
#46

OpenAI

GPT-4.1

4.6%
#47

Meta

Llama 4 Scout

4.3%
#48

Baidu

ERNIE 4.5 300B A47B

3.5%
#49

Amazon

Nova 2 Lite

3.0%