IFBench

58の多様で検証可能なドメイン外制約における指示追従の汎化性能を評価するベンチマークで、特定の出力要件に従うモデルの能力をテストする。

モデル: 150 現在、 451 合計
中央値: 43.0%
範囲: 12.1%–82.9%

スコア分布

最低: 12.1%
中央値: 43.0%
最高: 82.9%

リーダーボード

非推奨のモデルを含める

ランク	モデル	スコア
101	Nanbeige4.1-3B Nanbeige	35.4%
102	Qwen3 Coder Next Alibaba	35.2%
103	Jamba 1.7 Large AI21 Labs	35.2%
104	MiniCPM5-1B (Non-reasoning) OpenBMB	35.2%
105	Hermes 4 - Llama-3.1 405B (Non-reasoning) Nous Research	34.8%
106	Sarvam 105B (High) Sarvam	34.4%
107	INTELLECT-3 (Based on GLM-4.5-Air) Prime Intellect	34.0%
108	Gemma 4 E2B (Non-reasoning) Google	33.6%
109	EXAONE 4.0 32B (Non-reasoning) LG AI Research	33.5%
110	Qwen3.5 4B (Non-reasoning) Alibaba	33.3%
111	LFM2.5-VL-1.6B Liquid AI	33.1%
112	Mistral Small 4 (Non-reasoning) Mistral	32.8%
113	Olmo 3 7B Instruct Allen Institute for AI	32.8%
114	Hermes 4 - Llama-3.1 405B (Reasoning) Nous Research	32.7%
115	Ministral 3 14B Mistral	32.0%
116	NVIDIA Nemotron Nano 12B v2 VL (Reasoning) NVIDIA	31.9%
117	Granite 4.0 H Small IBM	31.5%
118	Qwen3.5 2B (Reasoning) Alibaba	31.5%
119	Jamba 1.7 Mini AI21 Labs	31.4%
120	Hermes 4 - Llama-3.1 70B (Reasoning) Nous Research	31.3%
121	Qwen3 Omni 30B A3B Instruct Alibaba	31.2%
122	Llama 3.2 Instruct 11B (Vision) Meta	30.4%
123	Reka Flash 3 Reka AI	30.4%
124	Nova Micro Amazon	29.4%
125	Ministral 3 8B Mistral	29.1%
126	Qwen3.5 2B (Non-reasoning) Alibaba	29.1%
127	Hermes 4 - Llama-3.1 70B (Non-reasoning) Nous Research	29.0%
128	Kimi Linear 48B A3B Instruct Kimi	28.1%
129	NVIDIA Nemotron Nano 9B V2 (Reasoning) NVIDIA	27.6%
130	NVIDIA Nemotron Nano 9B V2 (Non-reasoning) NVIDIA	27.1%
131	Molmo2-8B Allen Institute for AI	26.9%
132	Ministral 3 3B Mistral	26.8%
133	MiniCPM-V 4.6 1.3B OpenBMB	26.7%
134	Sarvam 30B (High) Sarvam	26.5%
135	Granite 4.0 H 1B IBM	26.2%
136	Apertus 70B Instruct Swiss AI Initiative	25.9%
137	NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning) NVIDIA	25.9%
138	Exaone 4.0 1.2B (Non-reasoning) LG AI Research	25.3%
139	Granite 4.0 Micro IBM	24.8%
140	Phi-4 Microsoft	23.5%
141	Exaone 4.0 1.2B (Reasoning) LG AI Research	23.0%
142	Apertus 8B Instruct Swiss AI Initiative	22.4%
143	Qwen3.5 0.8B (Non-reasoning) Alibaba	21.6%
144	Qwen3.5 0.8B (Reasoning) Alibaba	21.5%
145	Phi-4 Mini Instruct Microsoft	21.1%
146	Tiny Aya Global Cohere	20.1%
147	Molmo 7B-D Allen Institute for AI	19.7%
148	Granite 4.0 H 350M IBM	17.6%
149	Granite 4.0 350M IBM	15.9%
150	Gemma 3 270M Google	12.1%

101
Nanbeige4.1-3B
Nanbeige
35.4%
102
Qwen3 Coder Next

データ提供：Artificial Analysis

Command Palette

IFBench

スコア分布

リーダーボード