GPQA Diamond

이 벤치마크는?

대학원 수준의 전문 지식을 측정하는 다지선다형 벤치마크. 생물학, 화학, 물리학 분야 전문가도 어려워하는 문제들로 구성됩니다. 점수는 정답률(%)입니다.

출처: Artificial Analysis
순위모델
#1

Google

Gemini 3.1 Pro

94.1%
#2

OpenAI

GPT-5.4

92.0%
#3

Grok

Grok 4.20 (Reasoning)

91.1%
#4

Google

Gemini 3 Flash

89.8%
#5

Anthropic

Claude Opus 4.6

89.6%
#6

Meta AI

Muse Spark

88.4%
#7

Qwen

Qwen3.6 Plus

88.2%
#8

Moonshot AI

Kimi K2.5

87.9%
#9

MiniMax

MiniMax M2.7

87.4%
#10

Xiaomi

MiMo-V2-Pro

87.0%
#11

Anthropic

Claude Opus 4.5

86.6%
#12

Qwen

Qwen3.5 397B A17B

86.1%
#13

Google

Gemma 4 31B

85.7%
#14

Grok

Grok 4.1 Fast (Reasoning)

85.3%
#15

MiniMax

MiniMax M2.5

84.8%
#16

Google

Gemini 2.5 Pro

84.4%
#17

DeepSeek

DeepSeek V3.2

84.0%
#18

Anthropic

Claude Sonnet 4.5

83.4%
#19

OpenAI

GPT-5.4 Mini

82.3%
#20

Google

Gemini 3.1 Flash Lite

82.2%
#21

Z.ai

GLM-5

82.0%
#22

Baidu

ERNIE 4.5 300B A47B

81.1%
#23

Anthropic

Claude Opus 4.1

80.9%
#24

NVIDIA

Nemotron 3 Super

80.0%
#25

Anthropic

Claude Sonnet 4.6

79.7%
#26

Anthropic

Claude Opus 4

79.6%
#27

Google

Gemini 2.5 Flash

79.0%
#28

Anthropic

Claude Sonnet 4

77.7%
#29

Baidu

ERNIE 5.0 Thinking

77.7%
#30

Grok

Grok 4.20

77.6%
#31

OpenAI

GPT-5.4 Nano

76.1%
#32

OpenAI

GPT-5 Mini

68.7%
#33

OpenAI

GPT-5

68.6%
#34

Anthropic

Claude Haiku 4.5

67.2%
#35

OpenAI

GPT OSS 120B

67.2%
#36

Meta AI

Llama 4 Maverick

67.1%
#37

OpenAI

GPT-5 Nano

67.0%
#38

OpenAI

GPT-4.1

66.6%
#39

Google

Gemini 2.5 Flash Lite

65.1%
#40

Grok

Grok 4.1 Fast

63.7%
#41

Meituan

Longcat Flash Chat

63.6%
#42

Amazon

Nova 2 Lite

60.3%
#43

Meta AI

Llama 4 Scout

58.7%
#44

Mistral AI

Mistral Small 4

30.2%