IFBench

このベンチマークについて

形式・長さ・スタイルなど様々な制約を含む複雑な指示をどれだけ忠実に実行できるかを評価するベンチマーク。スコアは遵守率(%)です。

出典: Artificial Analysis
順位モデル
#1

Grok

Grok 4.20 (Reasoning)

81.2%
#2

Google

Gemini 3 Flash

78.0%
#3

Google

Gemini 3.1 Flash Lite

77.2%
#4

Google

Gemini 3.1 Pro

77.1%
#5

Meta AI

Muse Spark

75.9%
#6

MiniMax

MiniMax M2.7

75.7%
#7

Google

Gemma 4 31B

75.6%
#8

Qwen

Qwen3.6 Plus

75.2%
#9

ChatGPT

GPT-5.4

73.9%
#10

Z.ai

GLM-5

72.3%
#11

MiniMax

MiniMax M2.5

71.6%
#12

NVIDIA

Nemotron 3 Super

71.5%
#13

Moonshot AI

Kimi K2.5

70.2%
#14

Xiaomi

MiMo-V2-Pro

68.8%
#15

ChatGPT

GPT-5 Nano

65.9%
#16

ChatGPT

GPT-5.4 Mini

64.8%
#17

ChatGPT

GPT-5.4 Nano

64.4%
#18

DeepSeek

DeepSeek V3.2

60.7%
#19

ChatGPT

GPT OSS 120B

58.3%
#20

Anthropic

Claude Opus 4.5

58.0%
#21

Anthropic

Claude Sonnet 4.5

57.3%
#22

Anthropic

Claude Opus 4.1

55.4%
#23

Anthropic

Claude Sonnet 4

54.7%
#24

Anthropic

Claude Haiku 4.5

54.3%
#25

Anthropic

Claude Opus 4

53.7%
#26

Anthropic

Claude Opus 4.6

53.1%
#27

Grok

Grok 4.1 Fast (Reasoning)

52.7%
#28

Qwen

Qwen3.5 397B A17B

51.6%
#29

Google

Gemini 2.5 Flash

50.3%
#30

Grok

Grok 4.20

49.3%
#31

Google

Gemini 2.5 Pro

48.7%
#32

ChatGPT

GPT-5 Mini

45.6%
#33

ChatGPT

GPT-5

45.0%
#34

Meituan

Longcat Flash Chat

43.1%
#35

ChatGPT

GPT-4.1

43.0%
#36

Meta AI

Llama 4 Maverick

43.0%
#37

Anthropic

Claude Sonnet 4.6

42.4%
#38

Google

Gemini 2.5 Flash Lite

41.8%
#39

Baidu

ERNIE 5.0 Thinking

41.4%
#40

Amazon

Nova 2 Lite

40.5%
#41

Meta AI

Llama 4 Scout

39.5%
#42

Baidu

ERNIE 4.5 300B A47B

39.1%
#43

Grok

Grok 4.1 Fast

36.5%