benchgap
Instruction following

IFBench leaderboard

As of 2026-11-21, the highest measured score on IFBench is 89.2% by Flint. 149 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Flint89.2%measured
2MAI-Thinking-185.0%measured
3Qwen3.8 Max82.8%measured
4MiniMax M382.3%estimated ± 6.6 pp, low confidence
5Inkling-Small82.2%measured
6Nemotron 3 Ultra81.7%measured
7Sage-A1.581.6%measured
8Qwen3.8-Omni-Flash81.5%measured
9Grok 4.381.3%measured
10Qwen3.8-Flash-Next81.3%measured
11Beam 1.0 501B-A23B FP880.7%measured
12dots3-note Preview80.4%measured
13Solar Open 280.0%measured
14Inkling79.8%measured
15Beam 1.0 501B-A23B79.6%measured
16Qwen3.8-27B79.5%measured
17Granite 4.2 8B79.3%measured
18Beam 1.0 501B-A23B NVFP479.2%measured
19Ripple-65B Preview79.2%measured
20Qwen3.7 Max79.1%measured
21Qwen3.7 Plus79.1%measured
22MiMo-V2.5-Pro78.8%estimated ± 6.6 pp, medium confidence
23Granite 4.2 30B77.2%measured
24Mercury 2.577.0%measured
25Muse Glimmer 30B77.0%measured
26Mercury 376.8%measured
27Qwen3.5 397B76.5%measured
28Gemini 3.5 Flash76.3%measured
29GPT-5.2-Codex76.1%estimated ± 6.6 pp, medium confidence
30A.X K275.9%measured
31Qwen3.6 Plus75.8%measured
32Spark-X2.6-4B75.8%measured
33Gemini 3.1 Pro75.5%estimated ± 6.6 pp, medium confidence
34Qwen 3.6 Max (preview)75.0%estimated ± 6.6 pp, medium confidence
35DeepSeek V4 Pro74.9%estimated ± 6.6 pp, medium confidence
36LFM2.5-24B-A2B74.7%measured
37GLM-5.174.6%estimated ± 6.6 pp, medium confidence
38Ling 3.0 Flash74.5%measured
39Granite 4.2 3B74.3%measured
40Kimi K2.674.3%estimated ± 6.6 pp, medium confidence
41Agents-A174.3%estimated ± 9.2 pp, medium confidence
42Nemotron 3 Nano Omni 30B A3B74.2%measured
43GPT-5.4 nano74.2%estimated ± 6.6 pp, medium confidence
44GPT-5.574.2%estimated ± 6.6 pp, medium confidence
45Muse Spark74.2%estimated ± 6.6 pp, medium confidence
46Ternary Bonsai 2 27B74.0%measured
47MiniMax M2.774.0%estimated ± 6.6 pp, medium confidence
48Qwen3.5-122B-A10B74.0%estimated ± 6.6 pp, medium confidence
49Gemma 4 31B73.8%estimated ± 6.6 pp, medium confidence
50Qwen3.5-27B73.8%estimated ± 6.6 pp, medium confidence
51GPT-5.273.6%estimated ± 6.6 pp, medium confidence
52GPT-5.3 Codex73.6%estimated ± 6.6 pp, medium confidence
53Ling 3.0 Flash FP873.4%measured
54LFM3-1.7B73.0%measured
55Nemotron 3.5 Lightning 30B A3B NVFP472.9%measured
56o3-mini72.6%estimated ± 9.2 pp, medium confidence
57K-EXAONE 2.072.6%measured
58Command A+72.0%estimated ± 6.6 pp, medium confidence
59GPT-5.472.0%estimated ± 6.6 pp, medium confidence
60Gemma 4 12B71.5%estimated ± 6.6 pp, medium confidence
61Nemotron 3.5 Super VL 120B A12B NVFP471.5%measured
62GLM-5.271.3%estimated ± 6.6 pp, medium confidence
63GPT-5.4 mini71.3%estimated ± 6.6 pp, medium confidence
64GLM-5-Turbo71.2%estimated ± 6.6 pp, medium confidence
65GPT-571.1%estimated ± 6.6 pp, medium confidence
66GPT-5.170.9%estimated ± 6.6 pp, medium confidence
67Nemotron 3.5 Super VL 120B A12B70.9%measured
68GPT-5.6 Sol70.7%estimated ± 6.6 pp, medium confidence
69Qwen3.5-35B-A3B70.5%estimated ± 6.6 pp, medium confidence
70Gemma 4 26B A4B70.4%estimated ± 6.6 pp, medium confidence
71GLM-570.3%estimated ± 6.6 pp, medium confidence
72o369.4%estimated ± 6.6 pp, medium confidence
73GPT-5.6 Terra69.2%estimated ± 6.6 pp, medium confidence
74Agents-A1-4B69.1%measured
75Gemini 3 Pro68.4%estimated ± 6.6 pp, medium confidence
76o168.3%estimated ± 6.6 pp, medium confidence
77Kimi K2.568.2%estimated ± 6.6 pp, medium confidence
78GPT-5.1-Codex68.0%estimated ± 6.6 pp, medium confidence
79GPT-5.1-Codex-Max68.0%estimated ± 6.6 pp, medium confidence
80GPT-OSS 120B67.0%estimated ± 6.6 pp, medium confidence
81MiMo-V2-Pro66.8%estimated ± 6.6 pp, medium confidence
82Mistral Medium 3.5 128B66.8%estimated ± 6.6 pp, medium confidence
83o4-mini66.7%estimated ± 6.6 pp, medium confidence
84Mellum2.1-12B-A2.5B-Thinking66.3%estimated ± 9.2 pp, medium confidence
85MiniCPM5-2B66.3%measured
86GLM-4.766.0%estimated ± 6.6 pp, medium confidence
87Qwen3.6-27B65.7%estimated ± 6.6 pp, medium confidence
88Step 3.7 Flash65.5%estimated ± 6.6 pp, medium confidence
89GPT-OSS 20B63.7%estimated ± 6.6 pp, medium confidence
90K-Exaone63.3%estimated ± 6.6 pp, medium confidence
91Qwen3.6-35B-A3B63.1%estimated ± 6.6 pp, medium confidence
92Hy3 Preview63.1%measured
93Qwen3 235B 250762.5%estimated ± 9.2 pp, medium confidence
94Claude Fable 562.4%estimated ± 6.6 pp, medium confidence
95Kimi K2.7 Code62.1%estimated ± 6.6 pp, medium confidence
96Claude Opus 4.861.5%estimated ± 6.6 pp, medium confidence
97GLM-5V-Turbo60.8%estimated ± 6.6 pp, medium confidence
98Nemotron 3.5 Bolt E4B60.8%estimated ± 6.6 pp, medium confidence
99Claude Opus 4.759.3%estimated ± 6.6 pp, medium confidence
100LFM2.5-2.6B59.2%measured
101Claude Opus 4.5 Thinking59.0%estimated ± 6.6 pp, medium confidence
102North Mini Code58.8%estimated ± 6.6 pp, medium confidence
103Trinity-Large-Preview58.2%estimated ± 6.6 pp, medium confidence
104Claude Opus 4.558.0%measured
105Gemini 3 Flash57.7%estimated ± 6.6 pp, medium confidence
106Claude 4 Opus Thinking57.1%estimated ± 6.6 pp, medium confidence
107Grok 457.1%estimated ± 6.6 pp, medium confidence
108MiMo-V2-Omni57.0%estimated ± 6.6 pp, medium confidence
109Ling 2.6 Flash57.0%measured
110Claude Opus 4.6 (Adaptive)56.9%estimated ± 6.6 pp, medium confidence
111Grok 4.1 Fast (Reasoning)56.7%estimated ± 6.6 pp, medium confidence
112Qwen3.5-4B56.5%estimated ± 6.6 pp, medium confidence
113LFM2.5-8B-A1B56.5%measured
114OrcaCyber-Nano-1.056.3%measured
115Grok 4 Fast (Reasoning)56.0%estimated ± 6.6 pp, medium confidence
116Solar Pro 355.8%measured
117DeepSeek V3.255.6%estimated ± 6.6 pp, medium confidence
118Gemini 2.5 Pro55.6%estimated ± 6.6 pp, medium confidence
119Claude 3.7 Sonnet Thinking55.5%estimated ± 6.6 pp, medium confidence
120Mistral Small 455.5%estimated ± 6.6 pp, medium confidence
121Llama 3.3 70B55.2%estimated ± 6.6 pp, medium confidence
122Ultravox v0.6 Llama 3.3 70B55.2%estimated ± 6.6 pp, medium confidence
123Claude 4 Sonnet54.9%estimated ± 6.6 pp, medium confidence
124Qwen3 32B (Reasoning)54.9%estimated ± 9.2 pp, medium confidence
125Claude Opus 4.654.8%estimated ± 6.6 pp, medium confidence
126Gemma 4 E4B54.7%estimated ± 6.6 pp, medium confidence
127Qwen3 Max54.7%estimated ± 6.6 pp, medium confidence
128Phi-4-reasoning-plus54.7%estimated ± 9.2 pp, medium confidence
129Claude 3.7 Sonnet54.7%estimated ± 6.6 pp, medium confidence
130Qwen3-Omni-30B-A3B-Thinking54.6%estimated ± 6.6 pp, medium confidence
131Claude 4 Opus54.6%estimated ± 6.6 pp, medium confidence
132GPT-4.154.5%estimated ± 6.6 pp, medium confidence
133Llama 4 Maverick54.5%estimated ± 6.6 pp, medium confidence
134Claude 3.5 Haiku54.5%estimated ± 6.6 pp, low confidence
135Kimi K254.3%estimated ± 6.6 pp, low confidence
136Grok Code Fast 154.3%estimated ± 6.6 pp, low confidence
137Claude Sonnet 4.654.3%estimated ± 6.6 pp, low confidence
138DeepSeek V3 032454.3%estimated ± 6.6 pp, low confidence
139Gemini 2.0 Flash54.2%estimated ± 6.6 pp, low confidence
140MiMo-V2-Flash54.2%estimated ± 6.6 pp, low confidence
141DeepSeek-R154.1%estimated ± 6.6 pp, low confidence
142Llama 4 Scout54.1%estimated ± 6.6 pp, low confidence
143Mistral Medium 354.1%estimated ± 6.6 pp, low confidence
144Gemini 2.5 Flash54.1%estimated ± 6.6 pp, low confidence
145Llama 3.1 405B54.1%estimated ± 6.6 pp, low confidence
146QwQ-32B54.1%estimated ± 6.6 pp, low confidence
147GPT-4.1 mini54.0%estimated ± 6.6 pp, low confidence
148Nemotron Ultra 253B54.0%estimated ± 6.6 pp, low confidence
149Nova Pro54.0%estimated ± 6.6 pp, low confidence
150Gemma 4 E2B54.0%estimated ± 6.6 pp, low confidence
151DeepSeek V3.154.0%estimated ± 6.6 pp, low confidence
152GLM-4.5-Air54.0%estimated ± 6.6 pp, low confidence
153Nemotron 3 Nano 30B54.0%estimated ± 6.6 pp, low confidence
154GLM-4.653.9%estimated ± 6.6 pp, low confidence
155Grok 4.1 Fast53.9%estimated ± 6.6 pp, low confidence
156Mistral Large 353.9%estimated ± 6.6 pp, low confidence
157Claude 3 Haiku53.9%estimated ± 6.6 pp, low confidence
158DeepSeek V353.8%estimated ± 6.6 pp, low confidence
159Devstral Small53.8%estimated ± 6.6 pp, low confidence
160Sarvam 105B53.8%estimated ± 6.6 pp, low confidence
161GPT-4o53.8%estimated ± 6.6 pp, low confidence
162Solar Pro 253.8%estimated ± 6.6 pp, low confidence
163Exaone 4.0 32B53.8%estimated ± 6.6 pp, low confidence
164LFM2.5-VL-1.6B-Extract53.8%estimated ± 6.6 pp, low confidence
165GPT-4.1 nano53.7%estimated ± 6.6 pp, low confidence
166Gemma 3 27B53.7%estimated ± 6.6 pp, low confidence
167Gemini 2.5 Flash-Lite53.7%estimated ± 6.6 pp, low confidence
168Mistral Large 253.7%estimated ± 6.6 pp, low confidence
169Qwen3-Omni-30B-A3B-Instruct53.7%estimated ± 6.6 pp, low confidence
170GPT-4o mini53.7%estimated ± 6.6 pp, low confidence
171Mistral Small 3.153.7%estimated ± 6.6 pp, low confidence
172Llama 3.1 70B53.6%estimated ± 6.6 pp, low confidence
173Sarvam 30B53.6%estimated ± 6.6 pp, low confidence
174Granite-4.0-H-1B53.6%estimated ± 6.6 pp, low confidence
175Exaone 4.0 1.2B53.6%estimated ± 6.6 pp, low confidence
176Magistral Medium53.6%estimated ± 6.6 pp, low confidence
177Magistral Small53.6%estimated ± 6.6 pp, low confidence
178Phi-453.6%estimated ± 6.6 pp, low confidence
179Phi-4-reasoning53.6%estimated ± 6.6 pp, low confidence
180DeepSeek R1 Distill Qwen 32B53.6%estimated ± 6.6 pp, low confidence
181Granite-4.0-H-350M53.6%estimated ± 6.6 pp, low confidence
182Granite-4.0-350M53.5%estimated ± 6.6 pp, low confidence
183ZAYA1-8B52.6%measured
184Trinity-Large-Thinking52.3%measured
185Qwen3 235B-A22B (Reasoning)51.5%estimated ± 9.2 pp, medium confidence
186Qwen3 235B-A22B51.1%estimated ± 9.2 pp, medium confidence
187Qwen3 32B51.1%estimated ± 9.2 pp, medium confidence
188Ternary Bonsai 8B48.1%estimated ± 9.2 pp, medium confidence
189MiniCPM5-1B46.7%measured
190Celeris-146.0%estimated ± 9.2 pp, medium confidence
191Llama 3.1 8B45.2%estimated ± 9.2 pp, medium confidence
1921-bit Bonsai 8B43.9%estimated ± 9.2 pp, medium confidence
193Granite-4.0-1B38.9%estimated ± 9.2 pp, medium confidence
194LFM2.5-230M38.4%measured
195LFM2.5-350M38.0%estimated ± 9.2 pp, medium confidence
196Mellum2-12B-A2.5B-Thinking37.1%estimated ± 9.2 pp, medium confidence
197Mellum2-12B-A2.5B-Instruct35.7%estimated ± 9.2 pp, medium confidence
198Kanana-2 1.3B Instruct34.7%measured
199Kanana-2 3B Instruct33.3%measured
200Ternary Bonsai 4B28.7%estimated ± 9.2 pp, medium confidence
201LFM2.5-VL-3B25.8%measured
202Ternary Bonsai 1.7B25.1%estimated ± 9.2 pp, low confidence
203LLaDA2.2-mini24.9%measured
2041-bit Bonsai 4B24.3%estimated ± 9.2 pp, low confidence
2051-bit Bonsai 1.7B14.7%estimated ± 9.2 pp, low confidence
206LFM2.5-VL-450M12.6%estimated ± 9.2 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General