benchgap
Knowledge & reasoning

MMLU leaderboard

As of 2026-11-21, the highest measured score on MMLU is 91.8% by o1. 255 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.197.9%estimated ± 1.2 pp, medium confidence
2Claude Opus 597.3%estimated ± 1.2 pp, medium confidence
3Claude Fable 597.3%estimated ± 1.2 pp, medium confidence
4Gemini 3.1 Pro96.9%estimated ± 1.2 pp, medium confidence
5Gemini 3 Pro96.3%estimated ± 1.2 pp, medium confidence
6GPT-5.6 Sol95.6%estimated ± 1.2 pp, medium confidence
7Qwen3.8 Max95.2%estimated ± 1.2 pp, medium confidence
8GPT-5.594.8%estimated ± 1.2 pp, medium confidence
9Claude Sonnet 4.5 Thinking94.3%estimated ± 1.2 pp, medium confidence
10Grok 4.2093.6%estimated ± 1.2 pp, medium confidence
11Gemini 3.1 Flash-Lite93.5%estimated ± 1.2 pp, medium confidence
12GLM-5.3-Flash93.4%estimated ± 1.2 pp, medium confidence
13o393.1%estimated ± 1.2 pp, medium confidence
14Qwen3.5 Flash92.0%estimated ± 1.2 pp, medium confidence
15Qwen3.7 Plus91.9%estimated ± 2.1 pp, medium confidence
16o191.8%measured
17Qwen3.5 397B91.7%estimated ± 2.1 pp, medium confidence
18Qwen3.5-122B-A10B91.3%estimated ± 2.1 pp, medium confidence
19MiMo-V2.591.2%estimated ± 1.2 pp, high confidence
20Solar Open 291.1%estimated ± 2.1 pp, medium confidence
21DeepSeek-R190.8%measured
22GPT-4.590.8%measured
23Qwen3.6-35B-A3B90.7%estimated ± 2.1 pp, medium confidence
24DeepSeek-R1-052890.6%estimated ± 2.1 pp, medium confidence
25MAI-Thinking-190.6%estimated ± 2.1 pp, medium confidence
26Sarvam 105B90.6%measured
27Claude Opus 4.5 Thinking90.5%estimated ± 0.9 pp, medium confidence
28Claude Opus 4.690.5%estimated ± 0.9 pp, medium confidence
29Claude Opus 4.6 (Adaptive)90.5%estimated ± 0.9 pp, medium confidence
30Claude Opus 4.790.5%estimated ± 0.9 pp, medium confidence
31Claude Opus 4.890.5%estimated ± 0.9 pp, medium confidence
32Claude Sonnet 5.590.5%estimated ± 0.9 pp, medium confidence
33DeepSeek V4.1 Flash 101790.5%estimated ± 0.9 pp, medium confidence
34DeepSeek V4 Pro90.5%estimated ± 0.9 pp, medium confidence
35Gemini 3.5 Flash90.5%estimated ± 0.9 pp, medium confidence
36Gemini 3.6 Flash90.5%estimated ± 0.9 pp, medium confidence
37Gemini 3.7 Flash90.5%estimated ± 0.9 pp, medium confidence
38Gemini 3.8 Flash90.5%estimated ± 0.9 pp, medium confidence
39Gemini 3 Flash90.5%estimated ± 0.9 pp, medium confidence
40Gemini 4 Argon90.5%estimated ± 0.9 pp, medium confidence
41Gemini 4 Hydrogen90.5%estimated ± 0.9 pp, medium confidence
42GPT-5.290.5%estimated ± 0.9 pp, medium confidence
43GPT-5.490.5%estimated ± 0.9 pp, medium confidence
44GPT-5.6 Terra90.5%estimated ± 0.9 pp, medium confidence
45GPT-6.1 Terra90.5%estimated ± 0.9 pp, medium confidence
46GPT-6 Sol90.5%estimated ± 0.9 pp, medium confidence
47Grok 4.590.5%estimated ± 0.9 pp, medium confidence
48Grok 4.690.5%estimated ± 0.9 pp, medium confidence
49Kimi K3.1 Pro90.5%estimated ± 0.9 pp, medium confidence
50Kimi K390.5%estimated ± 0.9 pp, medium confidence
51MiniMax M390.5%estimated ± 0.9 pp, medium confidence
52MiniMax M3.1 Flash90.5%estimated ± 0.9 pp, medium confidence
53Muse Spark90.5%estimated ± 0.9 pp, medium confidence
54Muse Spark 1.190.5%estimated ± 0.9 pp, medium confidence
55Muse Spark 1.290.5%estimated ± 0.9 pp, medium confidence
56Qwen4 Max90.5%estimated ± 0.9 pp, medium confidence
57Muse Spark 1.390.5%estimated ± 0.9 pp, medium confidence
58GPT-5.6 Luna90.5%estimated ± 0.9 pp, medium confidence
59Grok 4.790.5%estimated ± 0.9 pp, medium confidence
60Muse Spark Open90.5%estimated ± 0.9 pp, medium confidence
61GPT-5.2-Codex90.5%estimated ± 0.9 pp, medium confidence
62Muse Wonder Full90.5%estimated ± 0.9 pp, medium confidence
63Claude Opus 4.590.5%estimated ± 0.9 pp, medium confidence
64Grok 490.5%estimated ± 0.9 pp, medium confidence
65DeepSeek V4 Flash 073190.5%estimated ± 0.9 pp, medium confidence
66GPT-590.5%estimated ± 0.9 pp, medium confidence
67Claude Sonnet 590.5%estimated ± 0.9 pp, medium confidence
68GPT-5.1-Codex90.5%estimated ± 0.9 pp, medium confidence
69GPT-5.1-Codex-Max90.5%estimated ± 0.9 pp, medium confidence
70GPT-6 Luna90.5%estimated ± 0.9 pp, medium confidence
71Inkling90.5%estimated ± 0.9 pp, medium confidence
72Gemini 2.5 Pro90.5%estimated ± 0.9 pp, medium confidence
73Claude Sonnet 4.690.5%estimated ± 0.9 pp, medium confidence
74Muse Wonder90.5%estimated ± 0.9 pp, medium confidence
75Ling 3.0 Tiny90.5%estimated ± 0.9 pp, medium confidence
76Qwen 3.6 Max (preview)90.5%estimated ± 0.9 pp, medium confidence
77GPT-5.190.5%estimated ± 0.9 pp, medium confidence
78GPT-5.4 mini90.5%estimated ± 0.9 pp, medium confidence
79Step 5 Preview90.5%estimated ± 0.9 pp, medium confidence
80Kimi K2.590.5%estimated ± 0.9 pp, medium confidence
81MiMo-V2.6-Pro90.5%estimated ± 0.9 pp, medium confidence
82Claude Haiku 5.590.5%estimated ± 0.9 pp, medium confidence
83Grok 4.390.5%estimated ± 0.9 pp, medium confidence
84Inkling 2 Small90.5%estimated ± 0.9 pp, high confidence
85Kimi K2.690.5%estimated ± 0.9 pp, high confidence
86GLM-5.390.5%estimated ± 0.9 pp, high confidence
87Inkling-Small90.5%estimated ± 0.9 pp, high confidence
88Hy3 Preview90.5%estimated ± 0.9 pp, high confidence
89Qwen3.7 Max90.5%estimated ± 0.9 pp, high confidence
90Beam 1.0 501B-A23B90.5%estimated ± 0.9 pp, high confidence
91Qwen3.8 Max Preview90.5%estimated ± 0.9 pp, high confidence
92GLM-4.790.5%estimated ± 0.9 pp, high confidence
93GLM-5V-Turbo90.5%estimated ± 0.9 pp, high confidence
94GLM-5-Turbo90.5%estimated ± 0.9 pp, high confidence
95Apodex 1.190.5%estimated ± 0.9 pp, high confidence
96Apodex 1.1 Mini90.5%estimated ± 0.9 pp, high confidence
97Claude 3.7 Sonnet90.5%estimated ± 0.9 pp, high confidence
98Claude 3.7 Sonnet Thinking90.5%estimated ± 0.9 pp, high confidence
99Muse Glimmer 30B90.5%estimated ± 0.9 pp, high confidence
100MiMo-V2.6-Flash90.5%estimated ± 0.9 pp, high confidence
101MiniMax M2.790.5%estimated ± 0.9 pp, high confidence
102MiMo-V2-Pro90.5%estimated ± 0.9 pp, high confidence
103Gemini 3.5 Flash-Lite90.5%estimated ± 0.9 pp, high confidence
104Qwen3.6 Plus90.5%estimated ± 0.9 pp, high confidence
105GLM-590.5%estimated ± 0.9 pp, high confidence
106Nemotron 3.5 Super VL 120B A12B90.5%estimated ± 0.9 pp, high confidence
107Gemini 2.5 Flash90.5%estimated ± 0.9 pp, high confidence
108Mistral Large 490.5%estimated ± 0.9 pp, high confidence
109Nemotron 3.5 Super VL 120B A12B NVFP490.5%estimated ± 0.9 pp, high confidence
110Step 3.7 Flash90.5%estimated ± 0.9 pp, high confidence
111GPT-5.4 nano90.5%estimated ± 0.9 pp, high confidence
112DeepSeek V390.5%estimated ± 0.9 pp, high confidence
113Grok 4.1 Fast (Reasoning)90.5%estimated ± 0.9 pp, high confidence
114Mistral Large 390.5%estimated ± 0.9 pp, high confidence
115Qwen3 Max90.5%estimated ± 0.9 pp, high confidence
116DeepSeek V3 032490.5%estimated ± 0.9 pp, high confidence
117GLM-5.290.5%estimated ± 0.9 pp, high confidence
118DeepSeek V3.290.4%estimated ± 0.9 pp, high confidence
119GLM-5.190.4%estimated ± 0.9 pp, high confidence
120Gemini 2.0 Flash90.4%estimated ± 0.9 pp, high confidence
121Grok Code Fast 190.4%estimated ± 0.9 pp, high confidence
122Qwen3.8-Flash-Next90.4%estimated ± 0.9 pp, high confidence
123DeepSeek V3.190.4%estimated ± 0.9 pp, high confidence
124Mistral Medium 3.5 128B90.4%estimated ± 0.9 pp, high confidence
125Grok 4 Fast (Reasoning)90.4%estimated ± 0.9 pp, high confidence
126Claude 4 Sonnet90.3%estimated ± 0.9 pp, high confidence
127MiMo-V2.5-Pro90.3%estimated ± 0.9 pp, high confidence
128GPT-4.190.2%measured
129Mercury 2.590.1%estimated ± 0.9 pp, high confidence
130Nemotron 3 Super 100B90.1%estimated ± 2.1 pp, high confidence
131Nemotron 3.5 Bolt 74B-A9B BF1690.0%estimated ± 0.9 pp, high confidence
132GLM-4.590.0%estimated ± 1.2 pp, high confidence
133GPT-OSS 120B90.0%measured
134o4-mini90.0%measured
135o3-pro89.9%estimated ± 3.8 pp, medium confidence
136Claude 4 Opus Thinking89.9%estimated ± 3.8 pp, medium confidence
137Claude 4.1 Opus89.9%estimated ± 3.8 pp, medium confidence
138Claude 4 Opus89.9%estimated ± 3.8 pp, high confidence
139Nemotron 3 Ultra89.9%estimated ± 0.9 pp, high confidence
140o1-pro89.9%estimated ± 3.8 pp, high confidence
141Qwen3 235B 250789.8%estimated ± 2.1 pp, high confidence
142GLM-4.689.7%estimated ± 0.9 pp, high confidence
143Mistral Small 489.7%estimated ± 0.9 pp, high confidence
144o1-preview89.7%estimated ± 3.8 pp, high confidence
145Ripple-65B Preview89.6%measured
146Kimi K289.5%measured
147Exaone 4.0 32B89.4%estimated ± 2.1 pp, high confidence
148Muse Ember89.3%estimated ± 0.9 pp, high confidence
149Grok 3 [Beta]88.6%estimated ± 2.1 pp, high confidence
150Qwen3.5-27B88.6%estimated ± 0.9 pp, high confidence
151MiniMax-Text-0188.5%measured
152Claude Haiku 4.588.4%estimated ± 1.2 pp, high confidence
153GPT-5.3 Codex88.2%estimated ± 8.3 pp, low confidence
154Hy388.2%estimated ± 8.3 pp, low confidence
155Kimi K2.7 Code88.2%estimated ± 8.3 pp, low confidence
156Ling 3.0 Flash VL88.2%estimated ± 8.3 pp, low confidence
157Claude 3.5 Sonnet88.2%estimated ± 1.2 pp, high confidence
158GPT-6.2 Astra88.1%estimated ± 4.2 pp, medium confidence
159GPT-6.2 Sol88.1%estimated ± 4.2 pp, medium confidence
160GPT-6 Astra88.1%estimated ± 4.2 pp, medium confidence
161Sakana Fugu88.1%estimated ± 4.2 pp, medium confidence
162Sakana Fugu-Ultra88.1%estimated ± 4.2 pp, medium confidence
163Claude Mythos 588.1%estimated ± 4.2 pp, medium confidence
164Ornith-1.5-397B88.1%estimated ± 4.2 pp, medium confidence
165Hy4 preview88.1%estimated ± 4.2 pp, medium confidence
166Qwen3.8-Omni-Flash88.1%estimated ± 4.2 pp, medium confidence
167Interfaze Beta88.1%estimated ± 4.2 pp, medium confidence
168Ornith-1.5-35B-A3B88.1%estimated ± 4.2 pp, medium confidence
169Ornith-1.5-9B88.1%estimated ± 4.2 pp, medium confidence
170Ternary Bonsai 2 27B88.1%estimated ± 4.2 pp, medium confidence
171Ling 3.0 Flash FP888.1%estimated ± 4.2 pp, medium confidence
172Claude Sonnet 4.588.1%estimated ± 4.2 pp, medium confidence
173Inkling 2 Mini88.1%estimated ± 0.9 pp, high confidence
174Grok 3 Mini88.1%estimated ± 4.2 pp, medium confidence
175Nemotron Ultra 253B88.0%estimated ± 4.2 pp, high confidence
176GLM-4.7-Flash88.0%estimated ± 4.2 pp, high confidence
177Qwen3.5-4B88.0%estimated ± 8.3 pp, medium confidence
178LFM2.5-24B-A2B88.0%estimated ± 4.2 pp, high confidence
179Trinity-Large-Thinking88.0%estimated ± 8.3 pp, medium confidence
180Qwen3 235B-A22B (Reasoning)88.0%estimated ± 4.2 pp, high confidence
181Spark-X2.6-4B87.9%estimated ± 4.2 pp, high confidence
182Nemotron 3 Nano 30B87.9%estimated ± 2.1 pp, high confidence
183Quasar 438B87.9%estimated ± 8.3 pp, medium confidence
184Qwen3 32B (Reasoning)87.9%estimated ± 4.2 pp, high confidence
185Qwen3-Omni-30B-A3B-Thinking87.9%estimated ± 8.3 pp, medium confidence
186DeepSeek R1 Distill Llama 70B87.8%estimated ± 4.2 pp, high confidence
187Qwen3 235B-A22B87.7%estimated ± 4.2 pp, high confidence
188Llama 4 Maverick87.7%estimated ± 8.3 pp, medium confidence
189DeepSeek-R1-0528-Qwen3-8B87.7%estimated ± 4.2 pp, high confidence
190Gemini 2.0 Flash-Lite87.6%estimated ± 3.8 pp, medium confidence
191DeepSeek R1 Distill Qwen 14B87.5%estimated ± 4.2 pp, high confidence
192Magistral Medium87.5%estimated ± 0.9 pp, high confidence
193GPT-4.1 mini87.5%measured
194Grok 287.5%measured
195Nemotron 3 Nano Omni 30B A3B87.5%estimated ± 2.1 pp, high confidence
196Mellum2-12B-A2.5B-Thinking87.4%estimated ± 4.2 pp, high confidence
197Qwen3-Omni-30B-A3B-Instruct87.4%estimated ± 8.3 pp, medium confidence
198Trinity-Large-Preview87.2%measured
199Qwen3 32B87.2%estimated ± 4.2 pp, high confidence
200Llama 4 Scout87.0%estimated ± 8.3 pp, medium confidence
201Qwen2.5-Max87.0%estimated ± 2.1 pp, high confidence
202Phi-4-reasoning-plus86.9%estimated ± 2.1 pp, high confidence
203o3-mini86.9%measured
204Celeris-186.9%estimated ± 2.1 pp, high confidence
205Qwen3.5-35B-A3B86.9%estimated ± 0.9 pp, high confidence
206LFM2.5-2.6B86.7%estimated ± 8.3 pp, medium confidence
207Gemini 1.5 Pro86.3%estimated ± 1.2 pp, high confidence
208Grok 2 mini86.2%measured
209Phi-4-reasoning86.2%estimated ± 2.1 pp, high confidence
210Gemma 4 31B86.2%estimated ± 0.9 pp, high confidence
211GPT-4o86.2%estimated ± 0.9 pp, high confidence
212Mistral Large 286.2%estimated ± 0.9 pp, high confidence
213ZAYA1-8B86.1%estimated ± 2.1 pp, high confidence
214Particle-186.1%measured
215Llama 3.1 70B86.0%measured
216Llama 3.3 70B86.0%measured
217Llama 3.1 405B85.9%estimated ± 8.3 pp, medium confidence
218QwQ-32B85.7%estimated ± 3.8 pp, high confidence
219Ultravox v0.6 Llama 3.3 70B85.5%estimated ± 8.3 pp, medium confidence
220LongCat-Flash-Lite-Sparse85.3%measured
221GPT-OSS 20B85.3%measured
222o1-mini85.2%measured
223Sarvam 30B85.1%measured
224Qwen3.6-27B85.0%estimated ± 0.9 pp, high confidence
225Phi-484.8%measured
226MiniCPM5-2B84.6%estimated ± 2.1 pp, high confidence
227MiMo-V2-Omni84.0%estimated ± 0.9 pp, high confidence
228Claude 3 Opus83.7%estimated ± 3.8 pp, high confidence
229Qwen2.5-VL-32B83.6%estimated ± 2.1 pp, high confidence
230DeepSeek R1 Distill Qwen 32B83.3%estimated ± 3.8 pp, high confidence
231ZAYA1-74B-Preview83.3%estimated ± 2.1 pp, high confidence
232Laguna XS.283.0%estimated ± 1.2 pp, high confidence
233Laguna M.182.9%estimated ± 1.2 pp, high confidence
234GPT-4 Turbo82.8%estimated ± 3.8 pp, high confidence
235Qwen2.5 Coder 32B Instruct82.8%estimated ± 3.8 pp, high confidence
236DeepSeek-V2.582.8%estimated ± 3.8 pp, high confidence
237Gemini 1.0 Pro82.8%estimated ± 3.8 pp, high confidence
238Mellum2-12B-A2.5B-Instruct82.7%estimated ± 4.2 pp, high confidence
239Exaone 4.0 1.2B82.3%estimated ± 8.3 pp, medium confidence
240GPT-4o mini82.0%measured
241Solar Pro 381.8%estimated ± 0.9 pp, high confidence
242Gemma 4 26B A4B81.6%estimated ± 0.9 pp, high confidence
243North Mini Code81.6%estimated ± 0.9 pp, high confidence
244Solar Pro 481.6%estimated ± 0.9 pp, high confidence
245Gemini 1.5 Flash81.5%estimated ± 1.2 pp, high confidence
246Mistral Medium 381.5%estimated ± 0.9 pp, high confidence
247Ling 3.0 Flash81.3%estimated ± 0.9 pp, high confidence
248Magistral Small81.1%estimated ± 0.9 pp, high confidence
249Claude 3 Haiku80.8%estimated ± 0.9 pp, high confidence
250A.X K280.7%estimated ± 0.9 pp, high confidence
251Qwen3.8-27B80.6%estimated ± 0.9 pp, high confidence
252Mistral Small 3.180.6%measured
253Grok 4.1 Fast80.6%estimated ± 0.9 pp, high confidence
254Nova Pro80.5%estimated ± 0.9 pp, high confidence
255K-Exaone80.4%estimated ± 0.9 pp, high confidence
256GLM-4.5-Air80.4%estimated ± 0.9 pp, high confidence
257Solar Pro 280.4%estimated ± 0.9 pp, high confidence
258Gemma 4 12B80.4%estimated ± 0.9 pp, high confidence
259Ling 2.6 Flash80.4%estimated ± 0.9 pp, high confidence
260MiMo-V2-Flash80.4%estimated ± 0.9 pp, high confidence
261Gemini 2.5 Flash-Lite80.4%estimated ± 0.9 pp, high confidence
262Devstral Small80.3%estimated ± 0.9 pp, high confidence
263Nemotron 3.5 Lightning 30B A3B NVFP480.3%estimated ± 0.9 pp, high confidence
264K-EXAONE 2.080.3%estimated ± 0.9 pp, high confidence
265Gemma 3 27B80.3%estimated ± 0.9 pp, medium confidence
266Granite 4.2 8B80.3%estimated ± 0.9 pp, medium confidence
267Granite 4.2 30B80.3%estimated ± 0.9 pp, medium confidence
268Command A+80.3%estimated ± 0.9 pp, medium confidence
269Gemma 4 E4B80.3%estimated ± 0.9 pp, medium confidence
270LFM2.5-8B-A1B80.3%estimated ± 0.9 pp, medium confidence
271Granite 4.2 3B80.3%estimated ± 0.9 pp, medium confidence
272LFM3-1.7B80.3%estimated ± 0.9 pp, medium confidence
273Claude 3.5 Haiku80.3%measured
274GPT-4.1 nano80.1%measured
275Gemma 4 E2B79.0%estimated ± 2.1 pp, high confidence
276Mixtral 8x22B Instruct v0.177.8%measured
277Gemma 2 27B76.2%measured
278Soofi S 30B-A3B73.8%estimated ± 2.1 pp, high confidence
279Llama 3.1 8B73.0%measured
280MiniCPM5-1B72.1%estimated ± 2.1 pp, high confidence
281Phi-3.5-mini69.0%measured
282Phi-4 Multimodal Instruct67.3%estimated ± 8.3 pp, medium confidence
283LFM2.5-VL-1.6B-Extract59.9%estimated ± 8.3 pp, medium confidence
284Granite-4.0-H-1B59.7%measured
285Granite-4.0-1B59.4%measured
286LFM2.5-230M45.4%estimated ± 2.1 pp, high confidence
287LFM2.5-350M45.1%estimated ± 2.1 pp, high confidence
288LFM2.5-VL-450M44.2%estimated ± 2.1 pp, high confidence
289Granite-4.0-H-350M36.2%measured
290Granite-4.0-350M35.0%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General