Best LLMs for math: open and proprietary

Open and proprietary models together on one 0–100 scale (mathematics). Switch to local models to see only what you can download and run.

40 benchmarks4 sourcesUpdated 3 Oct 2026Reference scale 2026-Q4How the score is built

Local models Overall

General Coding Agents Math Science Reasoning

Models ranked by math llmrun Score
#	Model	llmrun Score	Coding	Agents	Math	Science	Reasoning	Benchmarks	VRAM
1	GPT 6.1 Sol Max13 benchmarks		64	–	96	56	83	13	–
2	GPT 6 Astra Max19 benchmarks		83	72	96	57	86	19	–
3	Claude Opus 5.5 Max17 benchmarks		78	–	96	58	82	17	–
4	Claude Fable 5.1 Max15 benchmarks		79	–	95	57	75	15	–
5	Claude Fable 5 Max20 benchmarks		78	62	94	54	74	20	–
6	GPT 6 Sol Max18 benchmarks		79	–	94	56	75	18	–
7	Claude Sonnet 5.5 Max13 benchmarks		75	–	93	58	81	13	–
8	Claude Opus 5 Max22 benchmarks		78	72	91	55	75	22	–
9	GPT 5.6 Sol Max21 benchmarks		78	70	91	57	77	21	–
10	GPT 5.5 Pro7 benchmarks		–	–	90	–	70	7	–
11	GPT 5.6 Terra Max20 benchmarks		73	67	87	55	71	20	–
12	GPT 5.4 Pro (Mar 05, 2026)8 benchmarks		–	–	83	56	74	8	–
13	Claude Opus 4.8 Max19 benchmarks		64	55	82	48	66	19	–
14	GPT 5.6 Luna Max20 benchmarks		69	63	82	48	62	20	–
15	GPT 6 Luna Max18 benchmarks		69	–	82	47	60	18	–
16	Kimi K3open1674 GB · 26 benchmarks		73	61	81	52	63	26	1674 GB
17	GPT 5.4 (Mar 05, 2026)20 benchmarks		69	57	78	51	67	20	–
18	Qwen3.8 Max10 benchmarks		–	–	78	–	62	10	–
19	Muse Spark 1.3 Max6 benchmarks		–	–	78	54	61	6	–
20	GPT 5.2 Pro (Dec 11, 2025)3 benchmarks	—	–	–	77	–	–	3	–
21	Claude Sonnet 5 Max13 benchmarks		–	–	77	38	57	13	–
22	Muse Spark 1.36 benchmarks		–	–	76	–	53	6	–
23	DeepSeek V4.1 Flashopen458 GB · 10 benchmarks		54	–	76	45	57	10	458 GB
24	Gemini 3.7 Flash15 benchmarks		56	–	74	48	70	15	–
25	Claude Opus 4.7 Max15 benchmarks		67	–	73	42	64	15	–
26	GLM 5.3open456 GB · 21 benchmarks		67	68	72	49	58	21	456 GB
27	Qwen3.8 Max (Sep 02)6 benchmarks		–	–	72	–	–	6	–
28	DeepSeek V4 Pro 0813open991 GB · 21 benchmarks		63	–	71	46	65	21	991 GB
29	Claude Opus 4.6 Max12 benchmarks		–	–	71	45	55	12	–
30	DeepSeek V4 Flash 0731open183 GB · 21 benchmarks		59	–	71	45	62	21	183 GB
31	Claude Sonnet 4.6 Max11 benchmarks		–	–	70	34	59	11	–
32	Muse Spark 1.19 benchmarks		66	59	69	48	–	9	–
33	GLM 5.2open456 GB · 24 benchmarks		58	60	68	47	52	24	456 GB
34	Qwen3.7 Max17 benchmarks		59	–	68	43	58	17	–
35	Grok 4.516 benchmarks		65	–	66	47	60	16	–
36	Gemini 3.6 Flash16 benchmarks		52	–	66	45	63	16	–
37	GPT 5.2 (Dec 11, 2025)12 benchmarks		–	–	65	–	61	12	–
38	GPT 5 Pro (Oct 06, 2025)5 benchmarks		–	–	64	–	51	5	–
39	Grok 4.79 benchmarks		–	–	63	49	63	9	–
40	Kimi K2.6open620 GB · 18 benchmarks		57	–	63	43	51	18	620 GB
41	GPT 5 (Aug 07, 2025)20 benchmarks		58	–	62	40	48	20	–
42	GLM 5.3 Flashopen195 GB · 17 benchmarks		52	63	62	44	37	17	195 GB
43	Kimi K2.7 Codeopen620 GB · 19 benchmarks		53	46	61	40	50	19	620 GB
44	GPT 5.4 Mini (Mar 17, 2026)12 benchmarks		–	–	60	40	47	12	–
45	Gemini 3 Pro Preview17 benchmarks		62	55	60	46	54	17	–
46	Grok 4.20 0309 Reasoning8 benchmarks		–	–	58	–	49	8	–
47	Muse Spark6 benchmarks		–	–	58	47	–	6	–
48	DeepSeek V4 Proopen960 GB · 22 benchmarks		49	–	57	44	49	22	960 GB
49	Qwen3.8 27Bopen17.4 GB · 9 benchmarks		51	–	57	37	50	9	17.4 GB
50	GLM 5.1open457 GB · 13 benchmarks		53	54	57	40	46	13	457 GB
51	Grok 4.311 benchmarks		–	–	57	41	52	11	–
52	Inkling Smallopen160 GB · 12 benchmarks		–	–	57	41	54	12	160 GB
53	GPT 5 Mini (Aug 07, 2025)15 benchmarks		50	–	57	–	42	15	–
54	Qwen3.6 Plus15 benchmarks		47	–	55	38	41	15	–
55	Qwen3.6 27Bopen17.4 GB · 12 benchmarks		42	–	53	36	37	12	17.4 GB
56	Qwen3.5 397B A17Bopen266 GB · 9 benchmarks		–	–	52	–	46	9	266 GB
57	O4 Mini (Apr 16, 2025)19 benchmarks		49	–	52	34	42	19	–
58	Inklingopen572 GB · 21 benchmarks		43	–	52	40	53	21	572 GB
59	NVIDIA Nemotron 3 Ultra 550B A55B BF16open370 GB · 14 benchmarks		43	–	52	37	46	14	370 GB
60	O3 (Apr 16, 2025)19 benchmarks		52	–	52	35	46	19	–
61	GLM 4.7open216 GB · 13 benchmarks		41	32	51	36	35	13	216 GB
62	Gemini 2.5 Pro13 benchmarks		–	29	49	37	43	13	–
63	Gemini 3.1 Flash Lite6 benchmarks		–	–	49	–	38	6	–
64	Qwen3.6 35B A3Bopen22.0 GB · 12 benchmarks		44	–	49	35	38	12	22.0 GB
65	Qwen3.7 Flash5 benchmarks		–	–	49	–	46	5	–
66	GPT 5 Nano (Aug 07, 2025)15 benchmarks		42	–	48	–	25	15	–
67	Qwen3.6 Flash10 benchmarks		–	–	48	–	35	10	–
68	Grok 3 Mini Beta5 benchmarks		38	–	48	–	–	5	–
69	Qwen3.5 Flash10 benchmarks		–	–	47	–	43	10	–
70	O3 Mini (Jan 31, 2025)18 benchmarks		42	–	47	32	31	18	–
71	MiniMax M3open257 GB · 19 benchmarks		52	40	46	41	38	19	257 GB
72	Gemini 3.5 Flash Lite11 benchmarks		43	–	46	–	43	11	–
73	Qwen3 Max (Sep 23, 2025)13 benchmarks		–	–	46	–	36	13	–
74	GPT 5.5 Instant7 benchmarks		–	–	45	36	–	7	–
75	O1 (Dec 17, 2024)8 benchmarks		–	–	44	–	34	8	–
76	Claude Haiku 4.5 (Oct 01, 2025)8 benchmarks		–	–	44	–	40	8	–
77	DeepSeek R1 0528open415 GB · 11 benchmarks		46	–	44	–	32	11	415 GB
78	Grok 3 Beta6 benchmarks		–	–	40	–	–	6	–
79	DeepSeek R1open415 GB · 14 benchmarks		39	–	40	28	29	14	415 GB
80	DeepSeek R1 Distill Llama 70Bopen43.3 GB · 3 benchmarks	—	–	–	39	–	–	3	43.3 GB
81	DeepSeek R1 Distill Qwen 14Bopen9.6 GB · 4 benchmarks		–	–	38	–	–	4	9.6 GB
82	GPT 4.1 Mini (Apr 14, 2025)16 benchmarks		32	–	36	27	26	16	–
83	GPT 5.4 Nano 2026 03.176 benchmarks		–	–	36	–	27	6	–
84	O1 Mini (Sep 12, 2024)6 benchmarks		–	–	36	–	22	6	–
85	GPT 4.1 (Apr 14, 2025)17 benchmarks		39	–	34	26	26	17	–
86	GPT 4.5 Preview (Feb 27, 2025)11 benchmarks		36	–	33	27	28	11	–
87	DeepSeek v3 0324open415 GB · 13 benchmarks		39	–	33	27	24	13	415 GB
88	Mistral Medium 25055 benchmarks		–	–	33	23	–	5	–
89	Gemini 2.0 Flash 0017 benchmarks		–	–	32	–	24	7	–
90	O1 Preview (Sep 12, 2024)6 benchmarks		–	–	32	–	31	6	–
91	GPT 5.4 Mini 2026 03.176 benchmarks		–	–	30	–	29	6	–
92	GPT 4.1 Nano (Apr 14, 2025)13 benchmarks		21	–	30	17	17	13	–
93	Gemma 3 27B ITopen18.1 GB · 11 benchmarks		18	–	29	16	17	11	18.1 GB
94	Llama 4 Maverick 17B 128E Instructopen265 GB · 17 benchmarks		25	–	29	26	23	17	265 GB
95	Gemini 1.5 Pro 0029 benchmarks		–	–	28	22	21	9	–
96	Qwen Max (Jan 25, 2025)5 benchmarks		–	–	27	–	–	5	–
97	Qwen Plus (Jan 25, 2025)3 benchmarks	—	–	–	26	–	–	3	–
98	DeepSeek v3open415 GB · 10 benchmarks		37	–	26	22	21	10	415 GB
99	Phi 4open9.5 GB · 6 benchmarks		–	–	26	–	–	6	9.5 GB
100	Gemini 1.5 Flash 0026 benchmarks		–	–	25	–	–	6	–
101	Grok 2 (Dec 12)6 benchmarks		–	–	25	–	24	6	–
102	Qwen2.5 72B Instructopen44.6 GB · 9 benchmarks		–	–	25	–	23	9	44.6 GB
103	Llama 4 Scout 17B 16E Instructopen71.7 GB · 12 benchmarks		–	–	24	18	20	12	71.7 GB
104	Claude 3.5 Sonnet (Oct 22, 2024)11 benchmarks		38	26	23	21	29	11	–
105	Qwen2.5 32B Instructopen20.5 GB · 5 benchmarks		–	–	23	–	–	5	20.5 GB
106	Qwen Turbo (Nov 01, 2024)3 benchmarks	—	–	–	23	–	–	3	–
107	GPT 4o (Aug 06, 2024)8 benchmarks		–	–	22	–	19	8	–
108	GPT 4o Mini (Jul 18, 2024)14 benchmarks		18	–	22	–	17	14	–
109	Claude 3.5 Sonnet (Jun 20, 2024)7 benchmarks		–	–	22	–	26	7	–
110	Llama 3.1 405B Instructopen268 GB · 8 benchmarks		–	–	22	–	22	8	268 GB
111	GPT 4o (May 13, 2024)7 benchmarks		–	–	22	–	24	7	–
112	Mistral Large Instruct 2411open74.6 GB · 6 benchmarks		–	–	22	–	22	6	74.6 GB
113	GPT 4o (Nov 20, 2024)10 benchmarks		25	–	21	17	24	10	–
114	GPT 4 Turbo (Apr 09, 2024)8 benchmarks		–	–	21	–	23	8	–
115	Mistral Small 3.1 24B Instruct 2503open15.1 GB · 7 benchmarks		–	–	21	17	21	7	15.1 GB
116	Mistral Large Instruct 2407open80.9 GB · 7 benchmarks		–	–	20	–	22	7	80.9 GB
117	Claude 3.5 Haiku (Oct 22, 2024)9 benchmarks		29	–	20	13	–	9	–
118	Mistral Small 24B Instruct 2501open14.9 GB · 4 benchmarks		–	–	20	–	–	4	14.9 GB
119	Llama 3.1 Tulu 3 70B DPOopen43.3 GB · 3 benchmarks	—	–	–	19	–	–	3	43.3 GB
120	Llama 3.3 70B Instructopen46.6 GB · 13 benchmarks		22	–	19	17	21	13	46.6 GB

The range after each score is a 90% interval: given the boards a model has results on, its true score is very likely inside it. Short ranges mean many agreeing results; long ranges mean few or conflicting ones.

Why some models are missing. A model gets a score only with results on at least 4 benchmarks across at least 2 categories, one of them reasoning or coding. Skill scores need at least 2 benchmarks in that skill; otherwise they show as –.

VRAM is llmrun's estimate at Q4_K_M (or the smallest quantization we track) for the weights plus a working context. Proprietary models can't run locally, so the VRAM filter hides them.

How the llmrun Score is built · llmrun does not run these benchmarks; scores are aggregated from public sources.