Best LLM

What is the best LLM for math?

GPT-6 Astra and Gemini 4 Argon are practically level at the top in math.

Among the ten best, Claude Haiku 5.5 is the cheapest at $0.20 per million tokens.

The fastest among them is Claude Haiku 5.5, at 164 tokens per second.

Ranking: Math

Place and value in the area, plus overall index, price and speed. Higher values are better.

Rank	Model	Value	Overall index	Price (3:1)	Speed
1	GPT-6 AstraTop group	126.1	120.2	$20	58
2	GPT-6.1 SolUpper group	124.4	118.0	$4	44
3	Gemini 4 ArgonUpper group	118.9	121.4	$4	46
4	GPT-6 SolUpper group	117.7	109.1	$4	82
5	Claude Fable 5Upper group	117.6	112.8	$20	50
6	Claude Opus 5Middle group	113.8	117.4	$10	71
7	Muse WonderMiddle group	112.7	113.3	$8	n/a
8	Claude Haiku 5.5Middle group	107.2	102.8	$0.20	164
9	GPT-6 LunaMiddle group	106.0	95.8	$0.20	77
10	GPT-5.5Middle group	103.4	101.5	$11.25	87
11	Kimi K3Middle group	101.7	101.1	$6	61
12	Grok 4.6Middle group	101.4	102.1	$3	58
13	Gemini 3.7 FlashMiddle group	100.4	107.9	n/a	72
14	Gemini 3.8 FlashMiddle group	100.1	106.8	$1.50	81
15	GLM-5.3Middle group	95.6	100.6	$2.15	81
16	Grok 4.5Middle group	95.5	94.8	$3	52
17	Grok 4.7Middle group	94.3	103.3	$3	47
18	Gemini 3.6 FlashMiddle group	93.8	97.6	$1.50	110
19	Claude Sonnet 5Middle group	93.1	92.1	$4	77
20	Gemini 3.5 FlashMiddle group	92.7	97.8	n/a	142

Show all 46 ranked models

Rank	Model	Value	Overall index	Price (3:1)	Speed
21	GLM-5.2Middle group	91.9	89.0	$2.15	73
22	GLM-5.3-FlashMiddle group	91.2	94.3	$0.24	101
23	DeepSeek-V4-ProMiddle group	85.5	87.2	$0.99	44
24	GPT-5Lower group	75.6	77.5	$3.44	83
25	o4-miniLower group	67.7	63.5	n/a	n/a
26	GPT-5 miniLower group	66.9	65.8	$0.69	105
27	o3Lower group	66.7	74.2	$3.50	73
28	Qwen 3 235B A22BLower group	61.8	n/a	n/a	n/a
29	GPT OSS 120BLower group	55.6	43.4	$0.20	92
30	GLM-4.5Lower group	53.0	48.8	$1	28
31	Kimi K2 0711Lower group	52.1	n/a	$1	39
32	DeepSeek-R1Lower group	51.5	44.5	$1.15	n/a
33	o3-miniLower group	47.1	41.9	n/a	n/a
34	GPT-5 nanoLower group	42.5	32.4	$0.14	136
35	o1Lower group	42.3	49.2	n/a	n/a
36	GPT OSS 20BLower group	41.6	30.9	$0.06	85
37	DeepSeek-V3 0324Lower group	40.2	33.6	$0.44	39
38	MiniMax M2.1Lower group	40.2	57.3	$0.52	n/a
39	GPT-4.1Lower group	32.5	43.6	$3.50	81
40	GPT-4.1 miniLower group	29.3	32.3	$0.70	95
41	DeepSeek-V3Lower group	27.2	19.7	$0.37	40
42	Llama 4 ScoutLower group	10.0	4.7	$0.15	62
43	Command ALower group	1.3	0.4	$4.38	2
44	GPT-4o miniLower group	-2.9	-7.3	$0.26	66
45	Mistral Small 3.1Lower group	-7.7	-3.0	n/a	n/a
46	GPT-4.1 nanoLower group	-24.2	-15.8	n/a	110

Price in dollars per million tokens, mixed from 3 parts input and 1 part output. Speed in tokens per second.

The top 20: MathValue in the area · higher is better

GPT-6 AstraRank 1126.1

GPT-6.1 SolRank 2124.4

Gemini 4 ArgonRank 3118.9

GPT-6 SolRank 4117.7

Claude Fable 5Rank 5117.6

Claude Opus 5Rank 6113.8

Muse WonderRank 7112.7

Claude Haiku 5.5Rank 8107.2

GPT-6 LunaRank 9106.0

GPT-5.5Rank 10103.4

Kimi K3Rank 11101.7

Grok 4.6Rank 12101.4

Gemini 3.7 FlashRank 13100.4

Gemini 3.8 FlashRank 14100.1

GLM-5.3Rank 1595.6

Grok 4.5Rank 1695.5

Grok 4.7Rank 1794.3

Gemini 3.6 FlashRank 1893.8

Claude Sonnet 5Rank 1993.1

Gemini 3.5 FlashRank 2092.7

5 tests by independent evaluators feed this area. Each test counts with the result of the model in question.