Instruction following

IFEval leaderboard

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

As of 2026-10-07, the highest measured score on IFEval is 95.0% by Qwen3.5-27B. 135 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.5-27B	95.0%	measured
2	Agents-A1	94.8%	measured
3	Agents-A1-4B	94.8%	measured
4	Qwen3.7 Plus	94.6%	measured
5	Qwen3.6 Plus	94.3%	measured
6	Qwen3.7 Max	94.3%	measured
7	MiniMax M3	93.9%	estimated ± 1.3 pp, medium confidence
8	dots3-note Preview	93.9%	measured
9	Kimi K2.5	93.9%	measured
10	o3-mini	93.9%	measured
11	Nemotron 3 Ultra	93.9%	estimated ± 1.3 pp, medium confidence
12	Grok 4.3	93.9%	estimated ± 1.3 pp, medium confidence
13	MiMo-V2.5-Pro	93.8%	estimated ± 1.3 pp, high confidence
14	GPT-5.2-Codex	93.7%	estimated ± 1.3 pp, high confidence
15	Gemini 3.1 Pro	93.7%	estimated ± 1.3 pp, high confidence
16	Qwen 3.6 Max (preview)	93.7%	estimated ± 1.3 pp, high confidence
17	DeepSeek V4 Pro 0813	93.7%	estimated ± 1.3 pp, high confidence
18	Gemini 3.5 Flash	93.7%	estimated ± 1.3 pp, high confidence
19	GLM-5.1	93.7%	estimated ± 1.3 pp, high confidence
20	Kimi K2.6	93.7%	estimated ± 1.3 pp, high confidence
21	GPT-5.4 nano	93.7%	estimated ± 1.3 pp, high confidence
22	GPT-5.5	93.7%	estimated ± 1.3 pp, high confidence
23	Muse Spark	93.7%	estimated ± 1.3 pp, high confidence
24	MiniMax M2.7	93.7%	estimated ± 1.3 pp, high confidence
25	Gemma 4 31B	93.7%	estimated ± 1.3 pp, high confidence
26	GPT-5.2	93.7%	estimated ± 1.3 pp, high confidence
27	GPT-5.3 Codex	93.7%	estimated ± 1.3 pp, high confidence
28	Command A+	93.6%	estimated ± 1.3 pp, high confidence
29	GPT-5.4	93.6%	estimated ± 1.3 pp, high confidence
30	Gemma 4 12B	93.6%	estimated ± 1.3 pp, high confidence
31	GLM-5.2	93.6%	estimated ± 1.3 pp, high confidence
32	GPT-5.4 mini	93.6%	estimated ± 1.3 pp, high confidence
33	GLM-5-Turbo	93.6%	estimated ± 1.3 pp, high confidence
34	GPT-5 (high)	93.6%	estimated ± 1.3 pp, high confidence
35	GPT-5.1	93.6%	estimated ± 1.3 pp, high confidence
36	GPT-5.6 Sol	93.5%	estimated ± 1.3 pp, high confidence
37	Gemma 4 26B A4B	93.5%	estimated ± 1.3 pp, high confidence
38	Nemotron 3 Super 100B	93.5%	estimated ± 1.3 pp, high confidence
39	o3	93.5%	estimated ± 1.3 pp, high confidence
40	GPT-5.6 Terra	93.5%	estimated ± 1.3 pp, high confidence
41	Solar Pro 3	93.5%	estimated ± 1.3 pp, high confidence
42	Nemotron 3 Nano 30B	93.5%	estimated ± 1.3 pp, high confidence
43	GPT-5 (medium)	93.4%	estimated ± 1.3 pp, high confidence
44	Gemini 3 Pro	93.4%	estimated ± 1.3 pp, high confidence
45	Kimi K2.5 (Reasoning)	93.4%	estimated ± 1.3 pp, high confidence
46	GPT-5.1-Codex	93.4%	estimated ± 1.3 pp, high confidence
47	GPT-5.1-Codex-Max	93.4%	estimated ± 1.3 pp, high confidence
48	Qwen3.5-122B-A10B	93.4%	measured
49	GPT-OSS 120B	93.4%	estimated ± 1.3 pp, high confidence
50	MiMo-V2-Pro	93.3%	estimated ± 1.3 pp, high confidence
51	Mistral Medium 3.5 128B	93.3%	estimated ± 1.3 pp, high confidence
52	GLM-4.7	93.3%	estimated ± 1.3 pp, high confidence
53	Qwen3.6-27B	93.3%	estimated ± 1.3 pp, high confidence
54	Step 3.7 Flash	93.3%	estimated ± 1.3 pp, high confidence
55	GPT-OSS 20B	93.1%	estimated ± 1.3 pp, high confidence
56	K-Exaone	93.1%	estimated ± 1.3 pp, high confidence
57	Qwen3.6-35B-A3B	93.1%	estimated ± 1.3 pp, high confidence
58	Claude Fable 5	93.0%	estimated ± 1.3 pp, high confidence
59	Nemotron 3 Nano Omni 30B A3B	93.0%	estimated ± 1.3 pp, high confidence
60	Kimi K2.7 Code	93.0%	estimated ± 1.3 pp, high confidence
61	Claude Opus 4.8	92.9%	estimated ± 1.3 pp, high confidence
62	Inkling-Small	92.8%	estimated ± 3.5 pp, medium confidence
63	MAI-Thinking-1	92.8%	estimated ± 3.5 pp, medium confidence
64	Qwen3.8-Flash-Next	92.8%	estimated ± 3.5 pp, medium confidence
65	Qwen3.8 Max	92.8%	estimated ± 3.5 pp, medium confidence
66	Qwen3.8-Omni-Flash	92.8%	estimated ± 3.5 pp, medium confidence
67	Inkling	92.8%	estimated ± 3.5 pp, high confidence
68	Qwen3.8-27B	92.8%	estimated ± 3.5 pp, high confidence
69	Beam	92.8%	estimated ± 3.5 pp, high confidence
70	Solar Open 2	92.8%	estimated ± 3.5 pp, high confidence
71	Granite 4.2 8B	92.8%	estimated ± 3.5 pp, high confidence
72	Granite 4.2 30B	92.8%	estimated ± 3.5 pp, high confidence
73	Mercury 2.5	92.8%	estimated ± 3.5 pp, high confidence
74	Muse Glimmer 30B	92.8%	estimated ± 3.5 pp, high confidence
75	A.X K2	92.8%	estimated ± 3.5 pp, high confidence
76	Ling 3.0 Flash	92.8%	estimated ± 3.5 pp, high confidence
77	Granite 4.2 3B	92.8%	estimated ± 3.5 pp, high confidence
78	Ling 3.0 Flash FP8	92.8%	estimated ± 3.5 pp, high confidence
79	Nemotron 3.5 Lightning 30B A3B NVFP4	92.8%	estimated ± 3.5 pp, high confidence
80	GLM-5V-Turbo	92.8%	estimated ± 1.3 pp, high confidence
81	Hy3 Preview	92.7%	estimated ± 3.5 pp, high confidence
82	GLM-5	92.6%	measured
83	Qwen3.5 397B	92.6%	measured
84	Claude Opus 4.7 (Adaptive)	92.6%	estimated ± 1.3 pp, high confidence
85	Claude Opus 4.5 Thinking	92.5%	estimated ± 1.3 pp, high confidence
86	North Mini Code	92.5%	estimated ± 1.3 pp, high confidence
87	Ling 2.6 Flash	92.4%	estimated ± 1.3 pp, high confidence
88	K-EXAONE 2.0	92.4%	measured
89	Trinity-Large-Preview	92.3%	estimated ± 1.3 pp, high confidence
90	Trinity-Large-Thinking	92.3%	estimated ± 1.3 pp, high confidence
91	o1	92.2%	measured
92	Claude 4.1 Opus Thinking	92.2%	estimated ± 1.3 pp, high confidence
93	Gemini 3 Flash	92.2%	estimated ± 1.3 pp, high confidence
94	LFM2.5-2.6B	92.0%	estimated ± 3.5 pp, high confidence
95	Grok 4	92.0%	estimated ± 1.3 pp, high confidence
96	MiMo-V2-Omni	91.9%	estimated ± 1.3 pp, high confidence
97	Qwen3.5-35B-A3B	91.9%	measured
98	Claude Opus 4.6 (Adaptive)	91.9%	estimated ± 1.3 pp, high confidence
99	LFM2.5-8B-A1B	91.8%	measured
100	Grok 4.1 Fast (Reasoning)	91.8%	estimated ± 1.3 pp, high confidence
101	Qwen3.5 397B (Reasoning)	91.6%	estimated ± 1.3 pp, high confidence
102	Grok 4 Fast (Reasoning)	91.4%	estimated ± 1.3 pp, high confidence
103	Ternary Bonsai 2 27B	91.3%	measured
104	DeepSeek V3.2	91.2%	estimated ± 1.3 pp, high confidence
105	Gemini 2.5 Pro	91.1%	estimated ± 1.3 pp, high confidence
106	Mistral Small 4	91.0%	estimated ± 1.3 pp, high confidence
107	Mistral Small 4 (Reasoning)	91.0%	estimated ± 1.3 pp, high confidence
108	Claude Opus 4.5	90.9%	measured
109	Ultravox v0.6 Llama 3.3 70B	90.7%	estimated ± 1.3 pp, high confidence
110	Claude 4 Sonnet	90.3%	estimated ± 1.3 pp, high confidence
111	Claude Opus 4.6	90.1%	estimated ± 1.3 pp, high confidence
112	Gemma 4 E4B	90.0%	estimated ± 1.3 pp, high confidence
113	Qwen3 Max	90.0%	estimated ± 1.3 pp, high confidence
114	Claude Opus 4.7	89.8%	estimated ± 1.3 pp, high confidence
115	Qwen3-Omni-30B-A3B-Thinking	89.8%	estimated ± 1.3 pp, high confidence
116	Llama 4 Maverick	89.6%	estimated ± 1.3 pp, high confidence
117	DeepSeek V3.1 (Reasoning)	89.1%	estimated ± 1.3 pp, high confidence
118	Kimi K2	89.1%	estimated ± 1.3 pp, high confidence
119	Grok Code Fast 1	89.1%	estimated ± 1.3 pp, high confidence
120	Claude Sonnet 4.6	89.0%	estimated ± 1.3 pp, high confidence
121	DeepSeek V3 0324	88.9%	estimated ± 1.3 pp, high confidence
122	GPT-4.1 mini	88.5%	measured
123	MiMo-V2-Flash	88.5%	estimated ± 1.3 pp, high confidence
124	DeepSeek-R1	88.4%	estimated ± 1.3 pp, high confidence
125	Llama 4 Scout	88.3%	estimated ± 1.3 pp, high confidence
126	Mistral Medium 3	88.2%	estimated ± 1.3 pp, high confidence
127	Gemini 2.5 Flash	88.1%	estimated ± 1.3 pp, high confidence
128	Llama 3.1 405B	88.1%	estimated ± 1.3 pp, high confidence
129	Nemotron Ultra 253B	87.7%	estimated ± 1.3 pp, high confidence
130	Nova Pro	87.7%	estimated ± 1.3 pp, high confidence
131	Gemma 4 E2B	87.6%	estimated ± 1.3 pp, high confidence
132	DeepSeek V3.1	87.5%	estimated ± 1.3 pp, high confidence
133	GLM-4.5-Air	87.4%	estimated ± 1.3 pp, high confidence
134	GPT-4.1	87.4%	measured
135	GLM-4.6	86.9%	estimated ± 1.3 pp, high confidence
136	Grok 4.1 Fast	86.8%	estimated ± 1.3 pp, high confidence
137	MiniCPM5-2B	86.7%	measured
138	Mistral Large 3	86.6%	estimated ± 1.3 pp, high confidence
139	Claude 3 Haiku	86.6%	estimated ± 1.3 pp, high confidence
140	DeepSeek V3	86.1%	measured
141	ZAYA1-8B	85.6%	measured
142	Sarvam 105B	85.5%	estimated ± 1.3 pp, high confidence
143	GPT-4o	85.4%	estimated ± 1.3 pp, high confidence
144	Solar Pro 2	85.0%	estimated ± 1.3 pp, high confidence
145	Exaone 4.0 32B	84.8%	estimated ± 1.3 pp, high confidence
146	LFM2.5-VL-1.6B-Extract	84.5%	estimated ± 1.3 pp, high confidence
147	Gemma 3 27B	83.4%	estimated ± 1.3 pp, medium confidence
148	GPT-4.1 nano	83.2%	measured
149	Mistral Large 2	82.9%	estimated ± 1.3 pp, medium confidence
150	Qwen3-Omni-30B-A3B-Instruct	82.9%	estimated ± 1.3 pp, medium confidence
151	GPT-4o mini	82.7%	estimated ± 1.3 pp, medium confidence
152	LFM2.5-VL-3B	82.3%	measured
153	Kanana-2 3B Instruct	81.0%	measured
154	Celeris-1	80.8%	measured
155	MiniCPM5-1B	80.4%	measured
156	LLaDA2.2-mini	78.3%	estimated ± 3.5 pp, medium confidence
157	Kanana-2 1.3B Instruct	77.6%	measured
158	Sarvam 30B	77.1%	estimated ± 1.3 pp, medium confidence
159	Granite-4.0-H-1B	76.6%	estimated ± 1.3 pp, medium confidence
160	Mellum2-12B-A2.5B-Thinking	76.5%	measured
161	Mellum2-12B-A2.5B-Instruct	75.8%	measured
162	Exaone 4.0 1.2B	75.1%	estimated ± 1.3 pp, medium confidence
163	LFM2.5-230M	71.7%	measured
164	Phi-4	71.6%	estimated ± 1.3 pp, medium confidence
165	DeepSeek R1 Distill Qwen 32B	70.3%	estimated ± 1.3 pp, medium confidence
166	LFM2.5-VL-450M	61.2%	measured
167	Granite-4.0-H-350M	54.4%	estimated ± 1.3 pp, medium confidence
168	Granite-4.0-350M	47.5%	estimated ± 1.3 pp, medium confidence

IFEval leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General