Knowledge & reasoning

AA MMLU-Pro leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on AA MMLU-Pro is 89.8% by Gemini 3 Pro. 255 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Fable 5.1	91.6%	estimated ± 0.7 pp, medium confidence
2	Claude Opus 5.5	91.2%	estimated ± 0.7 pp, medium confidence
3	Claude Fable 5	91.1%	estimated ± 0.7 pp, medium confidence
4	GPT-6 Sol	91.1%	estimated ± 0.6 pp, medium confidence
5	GPT-5.5	90.7%	estimated ± 0.6 pp, medium confidence
6	GPT-6.1 Sol	90.6%	estimated ± 0.7 pp, medium confidence
7	GPT-6 Astra	90.5%	estimated ± 0.7 pp, medium confidence
8	Claude Opus 5	90.4%	estimated ± 0.7 pp, medium confidence
9	Gemini 3.7 Flash	90.4%	estimated ± 0.6 pp, medium confidence
10	Kimi K3.1 Pro	90.3%	estimated ± 0.6 pp, medium confidence
11	GPT-5.6 Sol	90.2%	estimated ± 0.7 pp, medium confidence
12	Grok 4.6	90.1%	estimated ± 0.6 pp, medium confidence
13	Gemini 4 Carbon	90.0%	estimated ± 0.7 pp, medium confidence
14	Grok 4.5	90.0%	estimated ± 0.6 pp, medium confidence
15	Kimi K3	90.0%	estimated ± 0.6 pp, medium confidence
16	Claude Opus 4.8	89.9%	estimated ± 0.6 pp, medium confidence
17	GPT-5.4	89.9%	estimated ± 0.6 pp, medium confidence
18	Claude Opus 4.7 (Adaptive)	89.8%	estimated ± 0.6 pp, medium confidence
19	Gemini 3 Pro	89.8%	measured
20	Muse Spark 1.1	89.5%	estimated ± 0.6 pp, medium confidence
21	Claude Opus 4.5 Thinking	89.5%	measured
22	DeepSeek V4.1 Flash 1017	89.4%	estimated ± 0.6 pp, medium confidence
23	Gemini 3.6 Flash	89.3%	estimated ± 0.6 pp, medium confidence
24	Muse Spark 1.2	89.3%	estimated ± 0.6 pp, medium confidence
25	Gemini 3.8 Flash	89.1%	estimated ± 0.6 pp, medium confidence
26	Muse Spark	89.0%	estimated ± 0.6 pp, medium confidence
27	Gemini 3.5 Flash	89.0%	estimated ± 0.7 pp, high confidence
28	Claude Sonnet 5.5	89.0%	estimated ± 0.7 pp, high confidence
29	Gemini 3.1 Pro	88.9%	estimated ± 1.5 pp, medium confidence
30	GPT-5.3 Codex	88.9%	estimated ± 1.5 pp, medium confidence
31	Ling 3.1 Flash	88.9%	estimated ± 1.5 pp, high confidence
32	Kimi K2.7 Code	88.9%	estimated ± 1.5 pp, high confidence
33	Apodex 1.1	88.9%	estimated ± 1.5 pp, high confidence
34	Apodex 1.1 Mini	88.9%	estimated ± 1.5 pp, high confidence
35	Hy3	88.9%	estimated ± 1.5 pp, high confidence
36	Muse Wonder	88.8%	estimated ± 0.6 pp, medium confidence
37	Claude Opus 4.6 (Adaptive)	88.8%	estimated ± 0.6 pp, medium confidence
38	MiMo-V2.6-Pro	88.7%	estimated ± 0.6 pp, medium confidence
39	Solar Pro 4	88.6%	estimated ± 1.5 pp, high confidence
40	DeepSeek V4 Pro 0813	88.6%	estimated ± 0.7 pp, high confidence
41	Claude Sonnet 5	88.6%	estimated ± 0.6 pp, medium confidence
42	Grok 4.7	88.5%	estimated ± 0.6 pp, medium confidence
43	A.X K2	88.5%	estimated ± 1.5 pp, high confidence
44	Gemini 4 Argon	88.4%	estimated ± 0.7 pp, high confidence
45	Claude Opus 4.7	88.4%	estimated ± 0.6 pp, medium confidence
46	Sage-A1.5	88.4%	estimated ± 1.5 pp, low confidence
47	Beam 1.0 501B-A23B FP8	88.4%	estimated ± 1.5 pp, low confidence
48	Beam 1.0 501B-A23B NVFP4	88.4%	estimated ± 1.5 pp, low confidence
49	GPT-6.1 Terra	88.3%	estimated ± 0.7 pp, high confidence
50	MiniMax M3.1 Flash	88.3%	estimated ± 0.6 pp, high confidence
51	DeepSeek V4 Flash 0731	88.2%	estimated ± 0.6 pp, medium confidence
52	GPT-5.6 Terra	88.2%	estimated ± 0.7 pp, high confidence
53	Claude Opus 4.6	88.1%	estimated ± 0.6 pp, high confidence
54	Claude 4.1 Opus Thinking	88.0%	measured
55	Grok 4.3	87.9%	estimated ± 0.6 pp, medium confidence
56	Hy3 Preview	87.9%	estimated ± 0.6 pp, medium confidence
57	Qwen3.5 397B (Reasoning)	87.8%	estimated ± 1.5 pp, high confidence
58	GPT-5.2-Codex	87.8%	estimated ± 0.6 pp, high confidence
59	Claude Haiku 5.5	87.8%	estimated ± 0.6 pp, medium confidence
60	Muse Spark 1.3	87.7%	estimated ± 0.7 pp, high confidence
61	GPT-5.6 Luna	87.6%	estimated ± 0.7 pp, high confidence
62	GPT-6 Luna	87.5%	estimated ± 0.6 pp, high confidence
63	Muse Spark Open	87.5%	estimated ± 0.6 pp, high confidence
64	GPT-5.2	87.4%	measured
65	Qwen 3.6 Max (preview)	87.4%	estimated ± 0.6 pp, high confidence
66	Gemini 3 Flash	87.3%	estimated ± 0.6 pp, high confidence
67	Muse Wonder Full	87.3%	estimated ± 0.7 pp, high confidence
68	Claude 4 Opus Thinking	87.3%	measured
69	Claude Opus 4.5	87.3%	estimated ± 0.6 pp, high confidence
70	MiniMax M3	87.3%	estimated ± 0.6 pp, high confidence
71	Step 5 Preview	87.2%	estimated ± 0.6 pp, high confidence
72	Sakana Fugu	87.2%	estimated ± 1.0 pp, low confidence
73	Sakana Fugu-Ultra	87.2%	estimated ± 1.0 pp, low confidence
74	GPT-5.1-Codex	87.2%	estimated ± 0.6 pp, high confidence
75	GPT-5.1-Codex-Max	87.2%	estimated ± 0.6 pp, high confidence
76	Claude Mythos 5	87.1%	estimated ± 1.0 pp, low confidence
77	Claude Sonnet 4.6	87.1%	estimated ± 0.6 pp, high confidence
78	GPT-5 (high)	87.1%	measured
79	GPT-5 (medium)	87.1%	estimated ± 0.6 pp, high confidence
80	GLM-5.3	87.1%	estimated ± 0.6 pp, medium confidence
81	Ornith-1.5-397B	87.0%	estimated ± 1.0 pp, low confidence
82	Qwen3.8 Max	87.0%	estimated ± 1.0 pp, low confidence
83	GPT-5.1	87.0%	measured
84	Hy4 preview	87.0%	estimated ± 1.0 pp, low confidence
85	GPT-5.4 mini	86.9%	estimated ± 0.6 pp, high confidence
86	Qwen3.8 Max Preview	86.9%	estimated ± 0.6 pp, medium confidence
87	Qwen3.8-Omni-Flash	86.9%	estimated ± 1.0 pp, low confidence
88	Qwen3.7 Plus	86.8%	estimated ± 1.0 pp, low confidence
89	o3-pro	86.8%	estimated ± 1.1 pp, high confidence
90	Ling 3.0 Tiny	86.8%	estimated ± 0.7 pp, high confidence
91	Interfaze Beta	86.8%	estimated ± 1.0 pp, low confidence
92	Ornith-1.5-35B-A3B	86.8%	estimated ± 1.0 pp, low confidence
93	Qwen3.7 Max	86.7%	estimated ± 0.6 pp, high confidence
94	Qwen3.5 397B	86.7%	estimated ± 1.0 pp, low confidence
95	Kimi K2.6	86.6%	estimated ± 0.6 pp, high confidence
96	Kimi K2.5	86.6%	estimated ± 0.6 pp, high confidence
97	Kimi K2.5 (Reasoning)	86.6%	estimated ± 0.6 pp, high confidence
98	Grok 4	86.6%	measured
99	Qwen3.5-122B-A10B	86.5%	estimated ± 1.0 pp, low confidence
100	Ornith-1.5-9B	86.5%	estimated ± 1.0 pp, low confidence
101	Qwen3.6-35B-A3B	86.5%	estimated ± 1.0 pp, low confidence
102	Ternary Bonsai 2 27B	86.5%	estimated ± 1.0 pp, low confidence
103	Gemma 4 31B	86.3%	estimated ± 1.0 pp, low confidence
104	MAI-Thinking-1	86.3%	estimated ± 1.0 pp, low confidence
105	GLM-5.2	86.3%	estimated ± 0.6 pp, medium confidence
106	Ling 3.0 Flash FP8	86.3%	estimated ± 1.0 pp, low confidence
107	Claude Sonnet 4.5	86.3%	estimated ± 1.0 pp, low confidence
108	Gemini 2.5 Pro	86.2%	measured
109	Gemma 4 26B A4B	86.1%	estimated ± 1.0 pp, low confidence
110	GLM-5-Turbo	86.1%	estimated ± 0.6 pp, high confidence
111	MiMo-V2-Pro	86.1%	estimated ± 0.6 pp, high confidence
112	Inkling	86.1%	estimated ± 0.6 pp, medium confidence
113	Claude 4.1 Opus	86.0%	estimated ± 1.1 pp, high confidence
114	Qwen3 235B 2507 (Reasoning)	86.0%	estimated ± 1.0 pp, low confidence
115	DeepSeek-R1-0528	86.0%	estimated ± 1.0 pp, low confidence
116	GLM-5	86.0%	estimated ± 0.6 pp, high confidence
117	MiMo-V2.6-Flash	86.0%	estimated ± 0.6 pp, high confidence
118	Grok 3 Mini	86.0%	estimated ± 1.0 pp, low confidence
119	GPT-OSS 120B	85.9%	estimated ± 1.0 pp, low confidence
120	GLM-5V-Turbo	85.9%	estimated ± 0.6 pp, high confidence
121	Qwen3.6 Plus	85.9%	estimated ± 0.6 pp, high confidence
122	Nemotron 3.5 Bolt 74B-A9B	85.9%	estimated ± 1.5 pp, high confidence
123	Ling 3.0 Flash VL	85.9%	estimated ± 1.5 pp, high confidence
124	Quasar 438B	85.9%	estimated ± 1.5 pp, high confidence
125	K-EXAONE 2.0	85.9%	estimated ± 1.5 pp, high confidence
126	Trinity-Large-Preview	85.9%	estimated ± 1.5 pp, high confidence
127	Trinity-Large-Thinking	85.9%	estimated ± 1.5 pp, high confidence
128	Celeris-1	85.9%	estimated ± 1.5 pp, medium confidence
129	Exaone 4.0 1.2B	85.9%	estimated ± 1.5 pp, medium confidence
130	Exaone 4.0 32B	85.9%	estimated ± 1.5 pp, medium confidence
131	Gemma 4 E2B	85.9%	estimated ± 1.5 pp, medium confidence
132	Granite-4.0-350M	85.9%	estimated ± 1.5 pp, medium confidence
133	Granite-4.0-H-1B	85.9%	estimated ± 1.5 pp, medium confidence
134	Granite-4.0-H-350M	85.9%	estimated ± 1.5 pp, medium confidence
135	LFM2.5-2.6B	85.9%	estimated ± 1.5 pp, medium confidence
136	LFM2.5-VL-1.6B-Extract	85.9%	estimated ± 1.5 pp, medium confidence
137	Llama 3.1 405B	85.9%	estimated ± 1.5 pp, medium confidence
138	Llama 3.1 70B	85.9%	estimated ± 1.5 pp, medium confidence
139	Llama 3.1 8B	85.9%	estimated ± 1.5 pp, medium confidence
140	Llama 3.3 70B	85.9%	estimated ± 1.5 pp, medium confidence
141	Llama 4 Maverick	85.9%	estimated ± 1.5 pp, medium confidence
142	Llama 4 Scout	85.9%	estimated ± 1.5 pp, medium confidence
143	MiniCPM5-2B	85.9%	estimated ± 1.5 pp, medium confidence
144	Mistral Small 4 (Reasoning)	85.9%	estimated ± 1.5 pp, medium confidence
145	North Mini Code	85.9%	estimated ± 1.5 pp, medium confidence
146	Phi-4 Multimodal Instruct	85.9%	estimated ± 1.5 pp, medium confidence
147	Qwen3.5-4B	85.9%	estimated ± 1.5 pp, medium confidence
148	Qwen3-Omni-30B-A3B-Instruct	85.9%	estimated ± 1.5 pp, medium confidence
149	Qwen3-Omni-30B-A3B-Thinking	85.9%	estimated ± 1.5 pp, medium confidence
150	Sarvam 105B	85.9%	estimated ± 1.5 pp, medium confidence
151	Sarvam 30B	85.9%	estimated ± 1.5 pp, medium confidence
152	Ultravox v0.6 Llama 3.3 70B	85.9%	estimated ± 1.5 pp, medium confidence
153	Nemotron 3 Super 100B	85.9%	estimated ± 1.0 pp, low confidence
154	o1-pro	85.8%	estimated ± 1.0 pp, low confidence
155	Gemma 4 12B	85.8%	estimated ± 1.0 pp, low confidence
156	o1	85.8%	estimated ± 0.6 pp, high confidence
157	Qwen3 235B 2507	85.7%	estimated ± 1.0 pp, low confidence
158	o3-mini	85.6%	estimated ± 1.0 pp, low confidence
159	GLM-4.7	85.6%	measured
160	MiniMax M2.7	85.6%	estimated ± 0.6 pp, high confidence
161	Beam 1.0 501B-A23B	85.6%	estimated ± 0.6 pp, high confidence
162	GLM-5.1	85.5%	estimated ± 0.6 pp, high confidence
163	Nemotron Ultra 253B	85.5%	estimated ± 1.0 pp, low confidence
164	Mistral Large 4	85.5%	estimated ± 0.6 pp, high confidence
165	Nemotron 3.5 Lightning 30B A3B NVFP4	85.5%	estimated ± 1.0 pp, low confidence
166	Grok 3 [Beta]	85.4%	estimated ± 1.0 pp, low confidence
167	GLM-4.7-Flash	85.4%	estimated ± 1.0 pp, low confidence
168	Grok 4.1 Fast (Reasoning)	85.4%	measured
169	Claude 4 Opus	85.4%	estimated ± 1.0 pp, low confidence
170	MiMo-V2.5-Pro	85.4%	estimated ± 0.6 pp, medium confidence
171	LFM2.5-24B-A2B	85.3%	estimated ± 1.0 pp, low confidence
172	o3	85.3%	measured
173	GPT-5.4 nano	85.3%	estimated ± 0.6 pp, high confidence
174	Claude 3.7 Sonnet Thinking	85.2%	estimated ± 0.6 pp, high confidence
175	Inkling-Small	85.2%	estimated ± 0.6 pp, medium confidence
176	Nemotron 3 Nano 30B	85.2%	estimated ± 1.0 pp, low confidence
177	Step 3.7 Flash	85.2%	estimated ± 0.6 pp, high confidence
178	DeepSeek V3.1 (Reasoning)	85.1%	measured
179	Nemotron 3 Nano Omni 30B A3B	85.1%	estimated ± 1.0 pp, low confidence
180	Claude 3.7 Sonnet	85.0%	estimated ± 0.6 pp, high confidence
181	Grok 4 Fast (Reasoning)	85.0%	measured
182	GPT-OSS 20B	85.0%	estimated ± 1.0 pp, low confidence
183	Ministral 3 14B (Reasoning)	85.0%	estimated ± 1.0 pp, low confidence
184	Qwen3 235B-A22B (Reasoning)	85.0%	estimated ± 1.0 pp, low confidence
185	ZAYA1-8B	84.9%	estimated ± 1.0 pp, low confidence
186	DeepSeek-R1	84.9%	measured
187	Qwen3.5-27B	84.9%	estimated ± 0.6 pp, medium confidence
188	MiMo-V2-Omni	84.8%	estimated ± 0.6 pp, medium confidence
189	Spark-X2.6-4B	84.8%	estimated ± 1.0 pp, low confidence
190	o4-mini	84.8%	estimated ± 0.6 pp, high confidence
191	GPT-4.5	84.8%	estimated ± 1.0 pp, low confidence
192	GPT-4.1	84.7%	estimated ± 0.6 pp, medium confidence
193	Kimi K2	84.7%	estimated ± 0.6 pp, medium confidence
194	Phi-4-reasoning-plus	84.7%	estimated ± 1.0 pp, low confidence
195	Nemotron 3.5 Super VL 120B A12B	84.6%	estimated ± 0.6 pp, high confidence
196	Qwen3 Max	84.6%	estimated ± 0.6 pp, high confidence
197	DeepSeek V3.2	84.6%	estimated ± 0.6 pp, high confidence
198	Qwen3.6-27B	84.6%	estimated ± 0.6 pp, medium confidence
199	Qwen3 32B (Reasoning)	84.6%	estimated ± 1.0 pp, low confidence
200	Nemotron 3.5 Super VL 120B A12B NVFP4	84.6%	estimated ± 0.6 pp, high confidence
201	Qwen3.8-Flash-Next	84.5%	estimated ± 0.7 pp, high confidence
202	Claude 4 Sonnet	84.5%	estimated ± 0.6 pp, medium confidence
203	Gemini 3.5 Flash-Lite	84.5%	estimated ± 0.6 pp, medium confidence
204	Qwen3.5-35B-A3B	84.5%	estimated ± 0.6 pp, medium confidence
205	Mistral Medium 3.5 128B	84.5%	estimated ± 0.7 pp, high confidence
206	Grok Code Fast 1	84.4%	estimated ± 0.6 pp, high confidence
207	Granite 4.2 30B	84.3%	estimated ± 1.0 pp, low confidence
208	o1-preview	84.3%	estimated ± 1.1 pp, medium confidence
209	Ling 3.0 Flash	84.3%	estimated ± 0.6 pp, medium confidence
210	Mercury 2.5	84.3%	estimated ± 0.7 pp, medium confidence
211	DeepSeek V3.1	84.3%	estimated ± 0.6 pp, high confidence
212	Nemotron 3.5 Bolt 74B-A9B BF16	84.3%	estimated ± 0.7 pp, medium confidence
213	Gemini 2.5 Flash	84.3%	estimated ± 0.6 pp, medium confidence
214	Phi-4-reasoning	84.3%	estimated ± 1.0 pp, low confidence
215	QwQ-32B	84.2%	estimated ± 1.0 pp, low confidence
216	Qwen3.8-27B	84.2%	estimated ± 0.6 pp, medium confidence
217	Mistral Small 4	84.2%	estimated ± 0.7 pp, medium confidence
218	GLM-4.6	84.2%	estimated ± 0.6 pp, medium confidence
219	Muse Glimmer 30B	84.2%	estimated ± 0.6 pp, medium confidence
220	Mistral Large 3	84.1%	estimated ± 0.6 pp, medium confidence
221	DeepSeek V3	84.1%	estimated ± 0.6 pp, medium confidence
222	DeepSeek V3 0324	84.1%	estimated ± 0.6 pp, medium confidence
223	Granite 4.2 8B	84.0%	estimated ± 1.0 pp, low confidence
224	Nemotron 3 Ultra	83.9%	estimated ± 0.6 pp, medium confidence
225	Gemini 2.0 Flash	83.9%	estimated ± 0.6 pp, medium confidence
226	Qwen3 235B-A22B	83.8%	estimated ± 1.0 pp, low confidence
227	Claude 3 Opus	83.7%	estimated ± 1.1 pp, medium confidence
228	MiMo-V2-Flash	83.6%	estimated ± 0.6 pp, medium confidence
229	GPT-4.1 mini	83.6%	estimated ± 0.6 pp, medium confidence
230	DeepSeek R1 Distill Qwen 32B	83.6%	estimated ± 1.1 pp, medium confidence
231	Command A+	83.6%	estimated ± 0.6 pp, medium confidence
232	K-Exaone	83.6%	estimated ± 0.6 pp, medium confidence
233	Magistral Medium	83.5%	estimated ± 0.6 pp, medium confidence
234	Gemini 1.5 Pro	83.5%	estimated ± 1.1 pp, medium confidence
235	Ling 2.6 Flash	83.5%	estimated ± 0.6 pp, medium confidence
236	GPT-4o	83.4%	estimated ± 0.6 pp, medium confidence
237	Qwen2.5-Max	83.4%	estimated ± 1.0 pp, low confidence
238	Grok 4.1 Fast	83.4%	estimated ± 0.6 pp, medium confidence
239	o1-mini	83.4%	estimated ± 1.0 pp, low confidence
240	Mistral Large 2	83.3%	estimated ± 0.6 pp, medium confidence
241	Gemini 1.5 Flash	83.3%	estimated ± 1.1 pp, medium confidence
242	Mistral Medium 3	83.3%	estimated ± 0.6 pp, medium confidence
243	GPT-4 Turbo	83.3%	estimated ± 1.1 pp, medium confidence
244	Claude 3.5 Sonnet	83.3%	estimated ± 1.0 pp, low confidence
245	GLM-4.5-Air	83.3%	estimated ± 0.6 pp, medium confidence
246	Qwen2.5 Coder 32B Instruct	83.2%	estimated ± 1.1 pp, medium confidence
247	Solar Pro 3	83.2%	estimated ± 0.6 pp, medium confidence
248	GPT-4o mini	83.2%	estimated ± 1.1 pp, medium confidence
249	DeepSeek-V2.5	83.2%	estimated ± 1.1 pp, medium confidence
250	Magistral Small	83.2%	estimated ± 0.6 pp, medium confidence
251	Gemma 4 E4B	83.1%	estimated ± 1.0 pp, low confidence
252	Mellum2-12B-A2.5B-Thinking	83.0%	estimated ± 1.0 pp, low confidence
253	Nova Pro	82.9%	estimated ± 0.6 pp, medium confidence
254	ZAYA1-74B-Preview	82.9%	estimated ± 1.0 pp, low confidence
255	Claude 3 Haiku	82.9%	estimated ± 0.6 pp, medium confidence
256	Gemini 1.0 Pro	82.9%	estimated ± 1.1 pp, medium confidence
257	Solar Pro 2	82.8%	estimated ± 0.6 pp, medium confidence
258	Mistral Small 3.1	82.7%	estimated ± 0.6 pp, medium confidence
259	Devstral Small	82.7%	estimated ± 0.6 pp, medium confidence
260	Gemini 2.5 Flash-Lite	82.7%	estimated ± 0.6 pp, medium confidence
261	Claude 3.5 Haiku	82.7%	estimated ± 0.6 pp, medium confidence
262	Phi-4	82.7%	estimated ± 1.0 pp, low confidence
263	Grok 2	82.7%	estimated ± 1.0 pp, low confidence
264	GPT-4.1 nano	82.6%	estimated ± 0.6 pp, medium confidence
265	Granite 4.2 3B	82.4%	estimated ± 1.0 pp, low confidence
266	Qwen3 32B	82.4%	estimated ± 1.0 pp, low confidence
267	LFM2.5-8B-A1B	82.3%	estimated ± 0.7 pp, medium confidence
268	Gemma 3 27B	82.3%	estimated ± 0.6 pp, medium confidence
269	LFM3-1.7B	82.0%	estimated ± 0.7 pp, medium confidence
270	Grok 2 mini	81.7%	estimated ± 1.0 pp, low confidence

AA MMLU-Pro leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General