Knowledge & reasoning

MMLU-Redux leaderboard

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

As of 2026-10-07, the highest measured score on MMLU-Redux is 96.6% by Claude Opus 4.5. 223 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	98.1%	estimated ± 1.1 pp, low confidence
2	Claude Opus 5.5	98.1%	estimated ± 1.1 pp, low confidence
3	Claude Fable 5	98.0%	estimated ± 1.1 pp, low confidence
4	GPT-6 Astra	97.9%	estimated ± 1.1 pp, low confidence
5	GPT-6.1 Sol	97.9%	estimated ± 1.1 pp, low confidence
6	Claude Opus 5	97.9%	estimated ± 1.1 pp, low confidence
7	GPT-5.6 Sol	97.8%	estimated ± 1.1 pp, low confidence
8	GPT-5.5	97.8%	estimated ± 1.1 pp, low confidence
9	Gemini 3 Pro	97.7%	estimated ± 1.1 pp, low confidence
10	Gemini 3.7 Flash	97.6%	estimated ± 1.1 pp, low confidence
11	Gemini 3.1 Pro	97.6%	estimated ± 1.1 pp, low confidence
12	Gemini 3.8 Flash	97.6%	estimated ± 1.1 pp, low confidence
13	GPT-6 Sol	97.6%	estimated ± 1.1 pp, low confidence
14	Claude Sonnet 5.5	97.6%	estimated ± 1.1 pp, low confidence
15	GPT-5.3 Codex	97.5%	estimated ± 1.1 pp, low confidence
16	Muse Spark 1.1	97.5%	estimated ± 1.1 pp, low confidence
17	Gemini 3.5 Flash	97.5%	estimated ± 1.1 pp, low confidence
18	Grok 4.5	97.5%	estimated ± 1.1 pp, low confidence
19	GPT-5.4	97.4%	estimated ± 1.1 pp, low confidence
20	Gemini 3.6 Flash	97.4%	estimated ± 1.1 pp, low confidence
21	Gemini 4 Argon	97.4%	estimated ± 1.1 pp, low confidence
22	Muse Spark	97.3%	estimated ± 1.1 pp, low confidence
23	Claude Opus 4.7 (Adaptive)	97.3%	estimated ± 1.1 pp, low confidence
24	Claude Opus 4.8	97.3%	estimated ± 1.1 pp, low confidence
25	Claude Mythos 5	97.3%	estimated ± 2.1 pp, low confidence
26	Grok 4.6	97.3%	estimated ± 1.1 pp, low confidence
27	Kimi K3	97.2%	estimated ± 1.1 pp, low confidence
28	Grok 4.7	97.2%	estimated ± 1.1 pp, low confidence
29	Claude Opus 4.6 (Adaptive)	97.2%	estimated ± 1.1 pp, low confidence
30	GPT-5.6 Terra	97.2%	estimated ± 1.1 pp, low confidence
31	Claude Opus 4.5 Thinking	97.2%	estimated ± 1.1 pp, low confidence
32	DeepSeek V4.1 Flash	97.1%	estimated ± 1.1 pp, low confidence
33	Gemini 3 Flash	97.1%	estimated ± 1.1 pp, low confidence
34	Muse Spark 1.2	97.1%	estimated ± 1.1 pp, low confidence
35	GPT-5.4 Pro	97.0%	estimated ± 2.1 pp, low confidence
36	Claude Opus 4.7	97.0%	estimated ± 1.1 pp, low confidence
37	GPT-5.2	97.0%	estimated ± 1.1 pp, low confidence
38	GPT-5.5 Pro	97.0%	estimated ± 2.1 pp, low confidence
39	GPT-6 Luna	97.0%	estimated ± 1.1 pp, low confidence
40	Muse Spark 1.3	96.9%	estimated ± 1.1 pp, low confidence
41	Hy4 preview	96.9%	estimated ± 2.1 pp, low confidence
42	GPT-5.6 Luna	96.9%	estimated ± 1.1 pp, low confidence
43	Inkling	96.8%	estimated ± 1.1 pp, low confidence
44	Step 5 Preview	96.8%	estimated ± 1.1 pp, low confidence
45	dots3-note Preview	96.8%	estimated ± 2.1 pp, low confidence
46	GPT-5.2-Codex	96.7%	estimated ± 1.1 pp, low confidence
47	Grok 4	96.7%	estimated ± 1.1 pp, medium confidence
48	GPT-5 (high)	96.7%	estimated ± 1.1 pp, medium confidence
49	Claude Sonnet 5	96.7%	estimated ± 1.1 pp, medium confidence
50	GPT-5.1-Codex	96.6%	estimated ± 1.1 pp, medium confidence
51	GPT-5.1-Codex-Max	96.6%	estimated ± 1.1 pp, medium confidence
52	Kimi K2.7 Code	96.6%	estimated ± 1.1 pp, medium confidence
53	GPT-5 (medium)	96.6%	estimated ± 1.1 pp, medium confidence
54	Claude Opus 4.5	96.6%	measured
55	Gemini 2.5 Pro	96.6%	estimated ± 1.1 pp, medium confidence
56	o3	96.5%	estimated ± 1.1 pp, medium confidence
57	Agents-A1	96.5%	estimated ± 2.1 pp, low confidence
58	GPT-5.1	96.4%	estimated ± 1.1 pp, medium confidence
59	GPT-5.4 mini	96.4%	estimated ± 1.1 pp, medium confidence
60	Ornith-1.5-397B	96.3%	estimated ± 2.1 pp, low confidence
61	Qwen3.8 Max	96.2%	estimated ± 2.1 pp, low confidence
62	MiMo-V2.6-Pro	96.1%	estimated ± 1.1 pp, medium confidence
63	Grok 4.3	96.1%	estimated ± 1.1 pp, medium confidence
64	o1	96.1%	estimated ± 1.1 pp, medium confidence
65	GLM-5.3	96.0%	estimated ± 1.1 pp, medium confidence
66	Inkling-Small	95.9%	estimated ± 1.1 pp, medium confidence
67	Kimi K2.6	95.9%	estimated ± 1.1 pp, medium confidence
68	Hy3	95.8%	estimated ± 1.1 pp, medium confidence
69	Apodex 1.1	95.7%	estimated ± 1.1 pp, medium confidence
70	Apodex 1.1 Mini	95.7%	estimated ± 1.1 pp, medium confidence
71	Qwen3.8 Max Preview	95.7%	estimated ± 1.1 pp, medium confidence
72	Hy3 Preview	95.7%	estimated ± 1.1 pp, medium confidence
73	Qwen 3.6 Max (preview)	95.6%	estimated ± 0.9 pp, medium confidence
74	DeepSeek-R1	95.6%	estimated ± 1.1 pp, medium confidence
75	Qwen3.8-Omni-Flash	95.6%	estimated ± 2.1 pp, medium confidence
76	Gemini 3.5 Flash-Lite	95.4%	estimated ± 1.1 pp, medium confidence
77	GLM-5V-Turbo	95.4%	estimated ± 1.1 pp, medium confidence
78	Ling 3.1 Flash	95.3%	estimated ± 1.1 pp, medium confidence
79	DeepSeek V3.1 (Reasoning)	95.3%	estimated ± 1.1 pp, medium confidence
80	GLM-5-Turbo	95.2%	estimated ± 1.1 pp, medium confidence
81	Sakana Fugu	95.2%	estimated ± 2.1 pp, medium confidence
82	Sakana Fugu-Ultra	95.2%	estimated ± 2.1 pp, medium confidence
83	Claude Haiku 5.5	95.2%	estimated ± 2.1 pp, low confidence
84	GLM-5.3-Flash	95.2%	estimated ± 2.1 pp, low confidence
85	GPT-4.1	95.1%	estimated ± 1.1 pp, medium confidence
86	Kimi K2	95.0%	estimated ± 1.1 pp, medium confidence
87	Qwen3.7 Max	95.0%	measured
88	MiMo-V2.6-Flash	95.0%	estimated ± 1.1 pp, medium confidence
89	Muse Glimmer 30B	95.0%	estimated ± 1.1 pp, medium confidence
90	MiniMax M2.7	94.9%	estimated ± 1.1 pp, medium confidence
91	Seed 2.1 Pro	94.9%	estimated ± 0.9 pp, high confidence
92	Qwen3.5 397B	94.9%	measured
93	MiMo-V2-Pro	94.9%	estimated ± 1.1 pp, medium confidence
94	Gemini 2.5 Flash	94.8%	estimated ± 1.1 pp, medium confidence
95	Claude 4.1 Opus Thinking	94.7%	estimated ± 2.1 pp, medium confidence
96	Mistral Large 4	94.7%	estimated ± 1.1 pp, medium confidence
97	Step 3.7 Flash	94.7%	estimated ± 1.1 pp, medium confidence
98	GPT-5.4 nano	94.7%	estimated ± 1.1 pp, medium confidence
99	o3-pro	94.6%	estimated ± 2.1 pp, medium confidence
100	Grok 4.1 Fast (Reasoning)	94.6%	estimated ± 1.1 pp, medium confidence
101	Mistral Large 3	94.6%	estimated ± 1.1 pp, medium confidence
102	Llama 4 Maverick	94.5%	estimated ± 1.1 pp, medium confidence
103	Mistral Medium 3.5 128B	94.5%	estimated ± 1.1 pp, medium confidence
104	Qwen3.6 Plus	94.5%	measured
105	Qwen3.7 Plus	94.5%	measured
106	Qwen3.5 397B (Reasoning)	94.5%	estimated ± 1.1 pp, medium confidence
107	Qwen3.8-Flash-Next	94.5%	estimated ± 1.1 pp, medium confidence
108	Qwen3 Max	94.4%	estimated ± 1.1 pp, medium confidence
109	DeepSeek V3 0324	94.4%	estimated ± 1.1 pp, medium confidence
110	GLM-5.2	94.4%	estimated ± 1.1 pp, medium confidence
111	Nemotron 3 Super 100B	94.4%	estimated ± 1.1 pp, medium confidence
112	DeepSeek V4 Pro 0813	94.4%	estimated ± 0.7 pp, high confidence
113	DeepSeek V3.2	94.3%	estimated ± 1.1 pp, medium confidence
114	GLM-5.1	94.3%	estimated ± 1.1 pp, medium confidence
115	Interfaze Beta	94.2%	estimated ± 2.1 pp, high confidence
116	Grok Code Fast 1	94.2%	estimated ± 1.1 pp, medium confidence
117	Kimi K2.5	94.2%	estimated ± 0.7 pp, high confidence
118	Kimi K2.5 (Reasoning)	94.2%	estimated ± 0.7 pp, high confidence
119	Llama 3.1 405B	94.2%	estimated ± 1.1 pp, medium confidence
120	Pareto 26.10 Preview	94.1%	estimated ± 3.0 pp, medium confidence
121	DeepSeek V3.1	94.1%	estimated ± 1.1 pp, medium confidence
122	Seed 2.1 Turbo	94.1%	estimated ± 0.9 pp, high confidence
123	Grok 4 Fast (Reasoning)	94.0%	estimated ± 1.1 pp, medium confidence
124	Claude 4 Sonnet	94.0%	estimated ± 1.1 pp, medium confidence
125	Nemotron 3 Ultra	94.0%	estimated ± 0.7 pp, high confidence
126	Trinity-Large-Preview	94.0%	estimated ± 1.1 pp, medium confidence
127	Trinity-Large-Thinking	94.0%	estimated ± 1.1 pp, medium confidence
128	Qwen3.5-122B-A10B	93.9%	estimated ± 0.7 pp, high confidence
129	MiMo-V2.5-Pro	93.9%	estimated ± 1.1 pp, medium confidence
130	Ornith-1.5-35B-A3B	93.9%	estimated ± 2.1 pp, medium confidence
131	Mercury 2.5	93.8%	estimated ± 1.1 pp, medium confidence
132	GPT-OSS 120B	93.8%	estimated ± 1.1 pp, medium confidence
133	Mistral Small 4	93.8%	estimated ± 1.1 pp, medium confidence
134	Mistral Small 4 (Reasoning)	93.8%	estimated ± 1.1 pp, medium confidence
135	Solar Pro 4	93.7%	estimated ± 0.7 pp, high confidence
136	DeepSeek V4 Flash 0731	93.7%	estimated ± 0.7 pp, high confidence
137	Solar Open 2	93.7%	estimated ± 0.7 pp, high confidence
138	GLM-4.6	93.7%	estimated ± 1.1 pp, medium confidence
139	Claude 4.1 Opus	93.7%	estimated ± 2.1 pp, medium confidence
140	Qwen3.5-27B	93.6%	estimated ± 0.7 pp, high confidence
141	Beam	93.5%	estimated ± 3.0 pp, medium confidence
142	Qwen3.6-27B	93.5%	measured
143	GLM-5	93.4%	estimated ± 0.7 pp, high confidence
144	GPT-4.1 mini	93.3%	estimated ± 1.1 pp, medium confidence
145	Nemotron Ultra 253B	93.3%	estimated ± 1.1 pp, medium confidence
146	GPT-4o	93.2%	estimated ± 1.1 pp, medium confidence
147	Mistral Large 2	93.2%	estimated ± 1.1 pp, medium confidence
148	Qwen3.5-35B-A3B	93.2%	estimated ± 0.7 pp, high confidence
149	Gemma 4 31B	93.1%	estimated ± 0.7 pp, high confidence
150	Qwen3.6-35B-A3B	93.1%	estimated ± 0.7 pp, high confidence
151	MAI-Thinking-1	93.0%	estimated ± 0.7 pp, high confidence
152	Claude Sonnet 4.5	93.0%	estimated ± 2.1 pp, high confidence
153	MiMo-V2-Omni	93.0%	estimated ± 1.1 pp, medium confidence
154	MiMo-V2-Flash	93.0%	estimated ± 0.7 pp, high confidence
155	Ultravox v0.6 Llama 3.3 70B	92.9%	estimated ± 1.1 pp, medium confidence
156	North Mini Code	92.9%	estimated ± 1.1 pp, medium confidence
157	Grok 4.20	92.8%	estimated ± 3.0 pp, medium confidence
158	A.X K2	92.7%	estimated ± 1.1 pp, medium confidence
159	Solar Pro 3	92.7%	estimated ± 1.1 pp, medium confidence
160	GLM-4.7	92.7%	estimated ± 0.7 pp, high confidence
161	Mistral Medium 3	92.6%	estimated ± 1.1 pp, medium confidence
162	Ling 3.0 Flash	92.6%	estimated ± 1.1 pp, medium confidence
163	Ling 3.0 Flash FP8	92.6%	estimated ± 1.1 pp, medium confidence
164	Ornith-1.5-9B	92.5%	estimated ± 2.1 pp, medium confidence
165	Claude 3 Haiku	92.3%	estimated ± 1.1 pp, medium confidence
166	Sarvam 105B	92.3%	estimated ± 1.1 pp, medium confidence
167	K-EXAONE 2.0	92.2%	estimated ± 0.7 pp, high confidence
168	Nemotron 3 Nano 30B	92.2%	estimated ± 1.1 pp, medium confidence
169	Grok 4.1 Fast	92.2%	estimated ± 1.1 pp, medium confidence
170	o1-pro	92.1%	estimated ± 2.1 pp, high confidence
171	Nova Pro	92.0%	estimated ± 1.1 pp, medium confidence
172	MiniMax M3	91.9%	estimated ± 1.1 pp, medium confidence
173	Qwen3 235B 2507	91.9%	estimated ± 0.7 pp, high confidence
174	K-Exaone	91.8%	estimated ± 1.1 pp, medium confidence
175	GLM-4.5-Air	91.8%	estimated ± 1.1 pp, medium confidence
176	Gemma 4 26B A4B	91.7%	estimated ± 0.7 pp, high confidence
177	Solar Pro 2	91.7%	estimated ± 1.1 pp, medium confidence
178	o3-mini	91.6%	estimated ± 2.1 pp, high confidence
179	GPT-OSS 20B	91.6%	estimated ± 1.1 pp, medium confidence
180	Ling 2.6 Flash	91.4%	estimated ± 1.1 pp, medium confidence
181	Qwen3.8-27B	91.4%	estimated ± 1.1 pp, medium confidence
182	Claude Opus 4.6	91.4%	estimated ± 0.7 pp, high confidence
183	Quasar 438B	91.4%	estimated ± 1.1 pp, medium confidence
184	Exaone 4.0 32B	91.3%	estimated ± 0.7 pp, high confidence
185	Llama 4 Scout	91.2%	estimated ± 1.1 pp, medium confidence
186	Nemotron 3.5 Lightning 30B A3B NVFP4	91.2%	estimated ± 0.7 pp, high confidence
187	Qwen3-Omni-30B-A3B-Thinking	90.9%	estimated ± 1.1 pp, medium confidence
188	Ling 3.0 Flash VL	90.7%	estimated ± 1.1 pp, medium confidence
189	Qwen3-Omni-30B-A3B-Instruct	90.7%	estimated ± 1.1 pp, medium confidence
190	Phi-4	90.6%	estimated ± 1.1 pp, medium confidence
191	GPT-4.1 nano	90.3%	estimated ± 1.1 pp, medium confidence
192	LongCat-Flash-Lite-Sparse	89.8%	estimated ± 0.7 pp, high confidence
193	Claude Sonnet 4.6	89.8%	estimated ± 0.7 pp, high confidence
194	Gemma 3 27B	89.8%	estimated ± 1.1 pp, medium confidence
195	Sarvam 30B	89.5%	estimated ± 1.1 pp, medium confidence
196	Ternary Bonsai 2 27B	89.1%	measured
197	Granite 4.2 30B	88.9%	estimated ± 0.7 pp, high confidence
198	Nemotron 3 Nano Omni 30B A3B	88.7%	estimated ± 0.7 pp, high confidence
199	Gemma 4 12B	88.7%	estimated ± 0.7 pp, high confidence
200	Celeris-1	87.9%	estimated ± 0.7 pp, high confidence
201	DeepSeek V3	87.9%	estimated ± 0.7 pp, high confidence
202	ZAYA1-8B	86.9%	estimated ± 0.7 pp, high confidence
203	Granite 4.2 8B	86.8%	estimated ± 0.7 pp, high confidence
204	Claude 3.5 Sonnet	86.6%	estimated ± 2.1 pp, high confidence
205	LFM2.5-8B-A1B	86.3%	estimated ± 1.1 pp, medium confidence
206	Mellum2-12B-A2.5B-Thinking	86.2%	measured
207	Command A+	85.7%	estimated ± 1.1 pp, medium confidence
208	Ling 3.0 Tiny	85.1%	estimated ± 1.1 pp, medium confidence
209	MiniCPM5-2B	84.7%	measured
210	Gemma 4 E4B	84.0%	estimated ± 0.7 pp, high confidence
211	ZAYA1-74B-Preview	83.2%	estimated ± 0.7 pp, high confidence
212	Granite 4.2 3B	83.0%	estimated ± 0.7 pp, high confidence
213	o1-preview	81.3%	estimated ± 2.1 pp, low confidence
214	LFM2.5-VL-1.6B-Extract	79.5%	estimated ± 1.1 pp, low confidence
215	Mellum2-12B-A2.5B-Instruct	78.1%	measured
216	LLaDA2.2-mini	78.1%	estimated ± 3.0 pp, medium confidence
217	Gemma 4 E2B	77.9%	estimated ± 0.7 pp, high confidence
218	Granite-4.0-H-1B	77.6%	estimated ± 1.1 pp, low confidence
219	Exaone 4.0 1.2B	76.9%	estimated ± 1.1 pp, low confidence
220	LFM2.5-2.6B	74.5%	estimated ± 1.1 pp, low confidence
221	Claude 3 Opus	73.7%	estimated ± 2.1 pp, low confidence
222	DeepSeek R1 Distill Qwen 32B	73.0%	estimated ± 2.1 pp, low confidence
223	Gemini 1.5 Pro	72.2%	estimated ± 2.1 pp, low confidence
224	Granite-4.0-350M	72.2%	estimated ± 1.1 pp, low confidence
225	Soofi S 30B-A3B	71.9%	estimated ± 0.7 pp, high confidence
226	Granite-4.0-H-350M	71.6%	estimated ± 1.1 pp, low confidence
227	GPT-4 Turbo	71.2%	estimated ± 2.1 pp, low confidence
228	Qwen2.5 Coder 32B Instruct	71.0%	estimated ± 2.1 pp, low confidence
229	GPT-4o mini	70.9%	estimated ± 2.1 pp, low confidence
230	Phi-4 Multimodal Instruct	70.5%	estimated ± 2.1 pp, low confidence
231	Gemini 1.0 Pro	70.4%	estimated ± 2.1 pp, low confidence
232	MiniCPM5-1B	70.1%	measured
233	LFM2.5-230M	44.9%	estimated ± 0.7 pp, medium confidence
234	LFM2.5-VL-450M	44.0%	estimated ± 0.7 pp, medium confidence

MMLU-Redux leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General