우리들의 비전

TAU2

이 벤치마크는?

도구 사용 능력을 평가하는 에이전트 벤치마크. 실제 사용자 작업을 시뮬레이션한 멀티스텝 태스크에서 AI의 도구 활용 및 계획 능력을 측정합니다. 점수는 태스크 성공률(%)입니다.

출처: Artificial Analysis
순위모델
#1

Z.ai

GLM-5

98.2%
#2

Z.ai

GLM-5.1

97.7%
#3

Alibaba

Qwen3.6 Plus

97.7%
#4

DeepSeek

DeepSeek V4 Pro

96.2%
#5

Moonshot AI

Kimi K2.5

95.9%
#6

Google

Gemini 3.1 Pro

95.6%
#7

MiniMax

MiniMax M2.5

95.3%
#8

DeepSeek

DeepSeek V4 Flash

95.0%
#9

Xiaomi

MiMo-V2-Pro

95.0%
#10

Grok

Grok 4.1 Fast (Reasoning)

93.3%
#11

Grok

Grok 4.20 (Reasoning)

93.0%
#12

Anthropic

Claude Opus 4.6

92.1%
#13

Meta

Muse Spark

91.5%
#14

DeepSeek

DeepSeek V3.2

90.6%
#15

Arcee AI

Trinity Large Thinking

90.1%
#16

Anthropic

Claude Opus 4.5

89.5%
#17

Anthropic

Claude Opus 4.7

88.6%
#18

OpenAI

GPT-5.4

87.1%
#19

MiniMax

MiniMax M2.7

84.8%
#20

Baidu

ERNIE 5.0 Thinking

83.9%
#21

Alibaba

Qwen3.5 397B A17B

83.9%
#22

Google

Gemini 3 Flash

80.4%
#23

Meituan

Longcat Flash Chat

79.5%
#24

Anthropic

Claude Sonnet 4.6

78.9%
#25

Anthropic

Claude Sonnet 4.5

78.1%
#26

Anthropic

Claude Opus 4

73.4%
#27

LG AI Research

K-EXAONE

73.2%
#28

Anthropic

Claude Opus 4.1

71.4%
#29

NVIDIA

Nemotron 3 Super

67.8%
#30

Anthropic

Claude Sonnet 4

64.6%
#31

Grok

Grok 4.1 Fast

63.7%
#32

Amazon

Nova 2 Lite

62.0%
#33

Google

Gemma 4 31B

59.9%
#34

Grok

Grok 4.20

59.9%
#35

Anthropic

Claude Haiku 4.5

54.7%
#36

Google

Gemini 2.5 Pro

54.1%
#37

OpenAI

GPT-5.4 Nano

52.6%
#38

OpenAI

GPT-4.1

47.1%
#39

OpenAI

GPT OSS 120B

45.0%
#40

Mistral AI

Mistral Small 4

41.2%
#41

OpenAI

GPT-5.4 Mini

36.5%
#42

OpenAI

GPT-5 Mini

31.9%
#43

Google

Gemini 2.5 Flash

31.6%
#44

Google

Gemini 3.1 Flash Lite

31.3%
#45

Google

Gemini 2.5 Flash Lite

30.4%
#46

OpenAI

GPT-5 Nano

30.4%
#47

Meta

Llama 4 Maverick

17.8%
#48

Meta

Llama 4 Scout

15.5%
#49

Baidu

ERNIE 4.5 300B A47B

0.0%
#50

OpenAI

GPT-5

0.0%