IFBench

Name: IFBench
Creator: aib vote
License: https://www.gnu.org/licenses/gpl-3.0.html

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。

順位	モデル
#1	Grok Grok 4.20 (Reasoning)	81.2%
#2	DeepSeek DeepSeek V4 Flash	79.2%
#3	Google Gemini 3 Flash	78.0%
#4	Google Gemini 3.1 Flash Lite	77.2%
#5	Google Gemini 3.1 Pro	77.1%
#6	DeepSeek DeepSeek V4 Pro	76.5%
#7	Z.ai GLM-5.1	76.3%
#8	Meta Muse Spark	75.9%
#9	MiniMax MiniMax M2.7	75.7%
#10	Google Gemma 4 31B	75.6%
#11	Alibaba Qwen3.6 Plus	75.2%
#12	OpenAI GPT-5.4	73.9%
#13	Z.ai GLM-5	72.3%
#14	MiniMax MiniMax M2.5	71.6%
#15	NVIDIA Nemotron 3 Super	71.5%
#16	Moonshot AI Kimi K2.5	70.2%
#17	Xiaomi MiMo-V2-Pro	68.8%
#18	LG AI Research K-EXAONE	67.3%
#19	OpenAI GPT-5 Nano	65.9%
#20	OpenAI GPT-5.4 Mini	64.8%
#21	OpenAI GPT-5.4 Nano	64.4%
#22	DeepSeek DeepSeek V3.2	60.7%
#23	Anthropic Claude Opus 4.7	58.6%
#24	OpenAI GPT OSS 120B	58.3%
#25	Anthropic Claude Opus 4.5	58.0%
#26	Anthropic Claude Sonnet 4.5	57.3%
#27	Arcee AI Trinity Large Thinking	56.3%
#28	Anthropic Claude Opus 4.1	55.4%
#29	Anthropic Claude Sonnet 4	54.7%
#30	Anthropic Claude Haiku 4.5	54.3%
#31	Anthropic Claude Opus 4	53.7%
#32	Anthropic Claude Opus 4.6	53.1%
#33	Grok Grok 4.1 Fast (Reasoning)	52.7%
#34	Alibaba Qwen3.5 397B A17B	51.6%
#35	Google Gemini 2.5 Flash	50.3%
#36	Grok Grok 4.20	49.3%
#37	Google Gemini 2.5 Pro	48.7%
#38	Mistral AI Mistral Small 4	48.2%
#39	OpenAI GPT-5 Mini	45.6%
#40	OpenAI GPT-5	45.0%
#41	Meituan Longcat Flash Chat	43.1%
#42	OpenAI GPT-4.1	43.0%
#43	Meta Llama 4 Maverick	43.0%
#44	Anthropic Claude Sonnet 4.6	42.4%
#45	Google Gemini 2.5 Flash Lite	41.8%
#46	Baidu ERNIE 5.0 Thinking	41.4%
#47	Amazon Nova 2 Lite	40.5%
#48	Meta Llama 4 Scout	39.5%
#49	Baidu ERNIE 4.5 300B A47B	39.1%
#50	Grok Grok 4.1 Fast	36.5%