Instruction following

IFBench leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on IFBench is 85.0% by MAI-Thinking-1. 125 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	MAI-Thinking-1	85.0%	measured
2	Qwen3.8 Max	82.8%	measured
3	MiniMax M3	82.3%	estimated ± 6.9 pp, low confidence
4	Inkling-Small	82.2%	measured
5	Nemotron 3 Ultra	81.7%	measured
6	Qwen3.8-Omni-Flash	81.5%	measured
7	Grok 4.3	81.3%	measured
8	Qwen3.8-Flash-Next	81.3%	measured
9	dots3-note Preview	80.4%	measured
10	Solar Open 2	80.0%	measured
11	Inkling	79.8%	measured
12	Beam	79.7%	measured
13	Qwen3.8-27B	79.5%	measured
14	Granite 4.2 8B	79.3%	measured
15	Qwen3.7 Max	79.1%	measured
16	Qwen3.7 Plus	79.1%	measured
17	MiMo-V2.5-Pro	78.9%	estimated ± 6.9 pp, medium confidence
18	Granite 4.2 30B	77.2%	measured
19	Mercury 2.5	77.0%	measured
20	Muse Glimmer 30B	77.0%	measured
21	GPT-5.2-Codex	76.3%	estimated ± 6.9 pp, medium confidence
22	Gemini 3.5 Flash	76.3%	measured
23	A.X K2	75.9%	measured
24	Qwen3.6 Plus	75.8%	measured
25	Gemini 3.1 Pro	75.8%	estimated ± 6.9 pp, medium confidence
26	Qwen 3.6 Max (preview)	75.2%	estimated ± 6.9 pp, medium confidence
27	DeepSeek V4 Pro 0813	75.1%	estimated ± 6.9 pp, medium confidence
28	GLM-5.1	74.9%	estimated ± 6.9 pp, medium confidence
29	Kimi K2.6	74.6%	estimated ± 6.9 pp, medium confidence
30	Ling 3.0 Flash	74.5%	measured
31	GPT-5.4 nano	74.5%	estimated ± 6.9 pp, medium confidence
32	GPT-5.5	74.5%	estimated ± 6.9 pp, medium confidence
33	Muse Spark	74.5%	estimated ± 6.9 pp, medium confidence
34	Granite 4.2 3B	74.3%	measured
35	MiniMax M2.7	74.3%	estimated ± 6.9 pp, medium confidence
36	Qwen3.5-122B-A10B	74.3%	estimated ± 6.9 pp, medium confidence
37	Nemotron 3 Nano Omni 30B A3B	74.2%	measured
38	Gemma 4 31B	74.2%	estimated ± 6.9 pp, medium confidence
39	Qwen3.5-27B	74.2%	estimated ± 6.9 pp, medium confidence
40	Ternary Bonsai 2 27B	74.0%	measured
41	GPT-5.2	74.0%	estimated ± 6.9 pp, medium confidence
42	GPT-5.3 Codex	74.0%	estimated ± 6.9 pp, medium confidence
43	Ling 3.0 Flash FP8	73.4%	measured
44	Nemotron 3.5 Lightning 30B A3B NVFP4	72.9%	measured
45	Agents-A1	72.8%	estimated ± 9.7 pp, low confidence
46	K-EXAONE 2.0	72.6%	measured
47	Command A+	72.4%	estimated ± 6.9 pp, medium confidence
48	GPT-5.4	72.4%	estimated ± 6.9 pp, medium confidence
49	Gemma 4 12B	72.0%	estimated ± 6.9 pp, medium confidence
50	GLM-5.2	71.8%	estimated ± 6.9 pp, medium confidence
51	GPT-5.4 mini	71.8%	estimated ± 6.9 pp, medium confidence
52	GLM-5-Turbo	71.7%	estimated ± 6.9 pp, medium confidence
53	GPT-5 (high)	71.6%	estimated ± 6.9 pp, medium confidence
54	GPT-5.1	71.4%	estimated ± 6.9 pp, medium confidence
55	GPT-5.6 Sol	71.2%	estimated ± 6.9 pp, medium confidence
56	o3-mini	71.1%	estimated ± 9.7 pp, medium confidence
57	Qwen3.5-35B-A3B	71.0%	estimated ± 6.9 pp, medium confidence
58	Gemma 4 26B A4B	70.9%	estimated ± 6.9 pp, medium confidence
59	GLM-5	70.8%	estimated ± 6.9 pp, medium confidence
60	Nemotron 3 Super 100B	70.0%	estimated ± 6.9 pp, medium confidence
61	o3	69.9%	estimated ± 6.9 pp, medium confidence
62	GPT-5.6 Terra	69.7%	estimated ± 6.9 pp, medium confidence
63	Nemotron 3 Nano 30B	69.6%	estimated ± 6.9 pp, medium confidence
64	GPT-5 (medium)	69.2%	estimated ± 6.9 pp, medium confidence
65	Agents-A1-4B	69.1%	measured
66	Gemini 3 Pro	69.0%	estimated ± 6.9 pp, medium confidence
67	o1	68.9%	estimated ± 6.9 pp, medium confidence
68	Kimi K2.5	68.8%	estimated ± 6.9 pp, medium confidence
69	Kimi K2.5 (Reasoning)	68.8%	estimated ± 6.9 pp, medium confidence
70	GPT-5.1-Codex	68.6%	estimated ± 6.9 pp, medium confidence
71	GPT-5.1-Codex-Max	68.6%	estimated ± 6.9 pp, medium confidence
72	GPT-OSS 120B	67.7%	estimated ± 6.9 pp, medium confidence
73	MiMo-V2-Pro	67.6%	estimated ± 6.9 pp, medium confidence
74	Mistral Medium 3.5 128B	67.6%	estimated ± 6.9 pp, medium confidence
75	GLM-4.7	66.8%	estimated ± 6.9 pp, medium confidence
76	Qwen3.6-27B	66.5%	estimated ± 6.9 pp, medium confidence
77	MiniCPM5-2B	66.3%	measured
78	Step 3.7 Flash	66.3%	estimated ± 6.9 pp, medium confidence
79	GPT-OSS 20B	64.6%	estimated ± 6.9 pp, medium confidence
80	K-Exaone	64.3%	estimated ± 6.9 pp, medium confidence
81	Qwen3.6-35B-A3B	64.1%	estimated ± 6.9 pp, medium confidence
82	Claude Fable 5	63.5%	estimated ± 6.9 pp, medium confidence
83	Kimi K2.7 Code	63.2%	estimated ± 6.9 pp, medium confidence
84	Hy3 Preview	63.1%	measured
85	Claude Opus 4.8	62.6%	estimated ± 6.9 pp, medium confidence
86	GLM-5V-Turbo	62.0%	estimated ± 6.9 pp, medium confidence
87	Claude Opus 4.7 (Adaptive)	60.6%	estimated ± 6.9 pp, medium confidence
88	Claude Opus 4.5 Thinking	60.3%	estimated ± 6.9 pp, medium confidence
89	North Mini Code	60.1%	estimated ± 6.9 pp, medium confidence
90	Trinity-Large-Preview	59.5%	estimated ± 6.9 pp, medium confidence
91	Trinity-Large-Thinking	59.5%	estimated ± 6.9 pp, medium confidence
92	Claude 4.1 Opus Thinking	59.2%	estimated ± 6.9 pp, medium confidence
93	LFM2.5-2.6B	59.2%	measured
94	Gemini 3 Flash	59.1%	estimated ± 6.9 pp, medium confidence
95	Grok 4	58.5%	estimated ± 6.9 pp, medium confidence
96	MiMo-V2-Omni	58.5%	estimated ± 6.9 pp, medium confidence
97	Claude Opus 4.6 (Adaptive)	58.3%	estimated ± 6.9 pp, medium confidence
98	Grok 4.1 Fast (Reasoning)	58.2%	estimated ± 6.9 pp, medium confidence
99	Claude Opus 4.5	58.0%	measured
100	Qwen3.5 397B	57.9%	estimated ± 6.9 pp, medium confidence
101	Qwen3.5 397B (Reasoning)	57.9%	estimated ± 6.9 pp, medium confidence
102	Grok 4 Fast (Reasoning)	57.6%	estimated ± 6.9 pp, medium confidence
103	DeepSeek V3.2	57.2%	estimated ± 6.9 pp, medium confidence
104	Gemini 2.5 Pro	57.1%	estimated ± 6.9 pp, medium confidence
105	Mistral Small 4	57.0%	estimated ± 6.9 pp, medium confidence
106	Mistral Small 4 (Reasoning)	57.0%	estimated ± 6.9 pp, medium confidence
107	Ling 2.6 Flash	57.0%	measured
108	Ultravox v0.6 Llama 3.3 70B	56.8%	estimated ± 6.9 pp, medium confidence
109	Claude 4 Sonnet	56.5%	estimated ± 6.9 pp, medium confidence
110	LFM2.5-8B-A1B	56.5%	measured
111	Claude Opus 4.6	56.4%	estimated ± 6.9 pp, medium confidence
112	Gemma 4 E4B	56.3%	estimated ± 6.9 pp, medium confidence
113	Qwen3 Max	56.3%	estimated ± 6.9 pp, medium confidence
114	Claude Opus 4.7	56.3%	estimated ± 6.9 pp, medium confidence
115	Qwen3-Omni-30B-A3B-Thinking	56.2%	estimated ± 6.9 pp, medium confidence
116	GPT-4.1	56.2%	estimated ± 6.9 pp, medium confidence
117	Llama 4 Maverick	56.2%	estimated ± 6.9 pp, medium confidence
118	DeepSeek V3.1 (Reasoning)	56.0%	estimated ± 6.9 pp, low confidence
119	Kimi K2	56.0%	estimated ± 6.9 pp, low confidence
120	Grok Code Fast 1	56.0%	estimated ± 6.9 pp, low confidence
121	Claude Sonnet 4.6	56.0%	estimated ± 6.9 pp, low confidence
122	DeepSeek V3 0324	56.0%	estimated ± 6.9 pp, low confidence
123	MiMo-V2-Flash	55.9%	estimated ± 6.9 pp, low confidence
124	DeepSeek-R1	55.8%	estimated ± 6.9 pp, low confidence
125	Llama 4 Scout	55.8%	estimated ± 6.9 pp, low confidence
126	Mistral Medium 3	55.8%	estimated ± 6.9 pp, low confidence
127	Gemini 2.5 Flash	55.8%	estimated ± 6.9 pp, low confidence
128	Llama 3.1 405B	55.8%	estimated ± 6.9 pp, low confidence
129	Solar Pro 3	55.8%	measured
130	GPT-4.1 mini	55.7%	estimated ± 6.9 pp, low confidence
131	Nemotron Ultra 253B	55.7%	estimated ± 6.9 pp, low confidence
132	Nova Pro	55.7%	estimated ± 6.9 pp, low confidence
133	Gemma 4 E2B	55.7%	estimated ± 6.9 pp, low confidence
134	DeepSeek V3.1	55.7%	estimated ± 6.9 pp, low confidence
135	GLM-4.5-Air	55.7%	estimated ± 6.9 pp, low confidence
136	GLM-4.6	55.6%	estimated ± 6.9 pp, low confidence
137	Grok 4.1 Fast	55.6%	estimated ± 6.9 pp, low confidence
138	Mistral Large 3	55.6%	estimated ± 6.9 pp, low confidence
139	Claude 3 Haiku	55.6%	estimated ± 6.9 pp, low confidence
140	DeepSeek V3	55.5%	estimated ± 6.9 pp, low confidence
141	Sarvam 105B	55.5%	estimated ± 6.9 pp, low confidence
142	GPT-4o	55.5%	estimated ± 6.9 pp, low confidence
143	Solar Pro 2	55.5%	estimated ± 6.9 pp, low confidence
144	Exaone 4.0 32B	55.5%	estimated ± 6.9 pp, low confidence
145	LFM2.5-VL-1.6B-Extract	55.5%	estimated ± 6.9 pp, low confidence
146	GPT-4.1 nano	55.4%	estimated ± 6.9 pp, low confidence
147	Gemma 3 27B	55.4%	estimated ± 6.9 pp, low confidence
148	Mistral Large 2	55.4%	estimated ± 6.9 pp, low confidence
149	Qwen3-Omni-30B-A3B-Instruct	55.4%	estimated ± 6.9 pp, low confidence
150	GPT-4o mini	55.4%	estimated ± 6.9 pp, low confidence
151	Sarvam 30B	55.3%	estimated ± 6.9 pp, low confidence
152	Granite-4.0-H-1B	55.3%	estimated ± 6.9 pp, low confidence
153	Exaone 4.0 1.2B	55.3%	estimated ± 6.9 pp, low confidence
154	Phi-4	55.3%	estimated ± 6.9 pp, low confidence
155	DeepSeek R1 Distill Qwen 32B	55.3%	estimated ± 6.9 pp, low confidence
156	Granite-4.0-H-350M	55.3%	estimated ± 6.9 pp, low confidence
157	Granite-4.0-350M	55.3%	estimated ± 6.9 pp, low confidence
158	ZAYA1-8B	52.6%	measured
159	MiniCPM5-1B	46.7%	measured
160	Celeris-1	44.6%	estimated ± 9.7 pp, medium confidence
161	LFM2.5-230M	38.4%	measured
162	Mellum2-12B-A2.5B-Thinking	35.8%	estimated ± 9.7 pp, medium confidence
163	Kanana-2 1.3B Instruct	34.7%	measured
164	Mellum2-12B-A2.5B-Instruct	34.4%	estimated ± 9.7 pp, medium confidence
165	Kanana-2 3B Instruct	33.3%	measured
166	LFM2.5-VL-3B	25.8%	measured
167	LLaDA2.2-mini	24.9%	measured
168	LFM2.5-VL-450M	12.0%	estimated ± 9.7 pp, low confidence

IFBench leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General