Ranggo ng bawat Mac para sa local LLM (M1 hanggang M6)

#	Mac	Maayos	Bandwidth	Pinakamalaki na maayos tumakbo	gpt-oss-20b
1	M5 Ultra · 512 GB	61	1,200 GB/s	DeepSeek V4.1 Flash 1017	284–505 tok/s
2	M3 Ultra · 512 GB	60	819 GB/s	DeepSeek V4.1 Flash 1017	194–345 tok/s
3	M5 Ultra · 256 GB	54	1,200 GB/s	MiMo V2.6 Flash	284–505 tok/s
4	M3 Ultra · 256 GB	53	819 GB/s	MiMo V2.6 Flash	194–345 tok/s
5	M2 Ultra · 192 GB	49	800 GB/s	MiniMax M2.7	189–337 tok/s
6	M2 Ultra · 128 GB	47	800 GB/s	Qwen3.5 122B A10B	189–337 tok/s
7	M1 Ultra · 128 GB	47	800 GB/s	Qwen3.5 122B A10B	189–337 tok/s
8	M5 Max · 128 GB	47	614 GB/s	Qwen3.5 122B A10B	145–259 tok/s
9	M4 Max · 128 GB	47	546 GB/s	Qwen3.5 122B A10B	129–230 tok/s
10	M3 Max · 128 GB	47	400 GB/s	Qwen3.5 122B A10B	95–168 tok/s
11	M5 Ultra · 96 GB	44	1,200 GB/s	gpt-oss-120b	284–505 tok/s
12	M3 Ultra · 96 GB	43	819 GB/s	gpt-oss-120b	194–345 tok/s
13	M2 Max · 96 GB	43	400 GB/s	gpt-oss-120b	95–168 tok/s
14	M2 Ultra · 64 GB	39	800 GB/s	Nemotron 3.5 Bolt 74B A9B	189–337 tok/s
15	M1 Ultra · 64 GB	39	800 GB/s	Nemotron 3.5 Bolt 74B A9B	189–337 tok/s
16	M5 Max · 64 GB	39	614 GB/s	Nemotron 3.5 Bolt 74B A9B	145–259 tok/s
17	M4 Max · 64 GB	39	546 GB/s	Nemotron 3.5 Bolt 74B A9B	129–230 tok/s
18	M3 Max · 64 GB	39	400 GB/s	Nemotron 3.5 Bolt 74B A9B	95–168 tok/s
19	M2 Max · 64 GB	39	400 GB/s	Nemotron 3.5 Bolt 74B A9B	95–168 tok/s
20	M1 Max · 64 GB	39	400 GB/s	Nemotron 3.5 Bolt 74B A9B	95–168 tok/s
21	M5 Max · 48 GB	38	614 GB/s	Qwen3.6 35B A3B	145–259 tok/s
22	M4 Max · 48 GB	38	546 GB/s	Qwen3.6 35B A3B	129–230 tok/s
23	M5 Max · 36 GB	38	460 GB/s	Qwen3.6 35B A3B	109–194 tok/s
24	M4 Max · 36 GB	38	410 GB/s	Qwen3.6 35B A3B	97–173 tok/s
25	M3 Max · 48 GB	38	400 GB/s	Qwen3.6 35B A3B	95–168 tok/s
26	M2 Max · 32 GB	38	400 GB/s	Qwen3.6 35B A3B	95–168 tok/s
27	M1 Max · 32 GB	38	400 GB/s	Qwen3.6 35B A3B	95–168 tok/s
28	M3 Max · 96 GB	38	300 GB/s	gpt-oss-120b	71–126 tok/s
29	M5 Pro · 64 GB	34	307 GB/s	Nemotron 3.5 Bolt 74B A9B	73–129 tok/s
30	M5 Pro · 48 GB	33	307 GB/s	Qwen3.6 35B A3B	73–129 tok/s
31	M3 Max · 36 GB	33	300 GB/s	Qwen3.6 35B A3B	71–126 tok/s
32	M4 Pro · 64 GB	31	273 GB/s	Nemotron 3.5 Bolt 74B A9B	65–115 tok/s
33	M4 Pro · 48 GB	30	273 GB/s	Qwen3.6 35B A3B	65–115 tok/s
34	M2 Pro · 32 GB	30	200 GB/s	Qwen3.6 35B A3B	47–84 tok/s
35	M1 Pro · 32 GB	30	200 GB/s	Qwen3.6 35B A3B	47–84 tok/s
36	M6 · 32 GB	29	170 GB/s	Qwen3.6 35B A3B	40–72 tok/s
37	M5 Pro · 24 GB	28	307 GB/s	Gemma 4 26B A4B	73–129 tok/s
38	M5 · 32 GB	28	153 GB/s	Qwen3.6 35B A3B	36–64 tok/s
39	M3 Pro · 36 GB	28	150 GB/s	Qwen3.6 35B A3B	36–63 tok/s
40	M4 · 32 GB	27	120 GB/s	Qwen3.6 35B A3B	28–51 tok/s
41	M4 Pro · 24 GB	25	273 GB/s	Gemma 4 26B A4B	65–115 tok/s
42	M6 · 24 GB	24	170 GB/s	Gemma 4 26B A4B	40–72 tok/s
43	M5 · 24 GB	23	153 GB/s	Gemma 4 26B A4B	36–64 tok/s
44	M2 Pro · 16 GB	22	200 GB/s	Phi-4 Reasoning Vision 15B	56–100 tok/s
45	M1 Pro · 16 GB	22	200 GB/s	Phi-4 Reasoning Vision 15B	56–100 tok/s
46	M4 · 24 GB	22	120 GB/s	Gemma 4 26B A4B	28–51 tok/s
47	M6 · 16 GB	20	153 GB/s	Gemma 4 12B Unified	43–77 tok/s
48	M5 · 16 GB	20	153 GB/s	Gemma 4 12B Unified	43–77 tok/s
49	M3 Pro · 18 GB	20	150 GB/s	Gemma 4 12B Unified	36–63 tok/s
50	M3 · 24 GB	20	100 GB/s	Gemma 4 26B A4B	24–42 tok/s
51	M2 · 24 GB	20	100 GB/s	Gemma 4 26B A4B	24–42 tok/s
52	M4 · 16 GB	19	120 GB/s	Qwen3.5 9B	34–60 tok/s
53	M3 · 16 GB	17	100 GB/s	Qwen3 8B	28–50 tok/s
54	M2 · 16 GB	17	100 GB/s	Qwen3 8B	28–50 tok/s
55	M3 · 8 GB	16	100 GB/s	Granite 4.2 8B	masyadong malaki
56	M2 · 8 GB	16	100 GB/s	Granite 4.2 8B	masyadong malaki
57	M1 · 16 GB	14	68 GB/s	Gemma 4 E4B	19–34 tok/s
58	M1 · 8 GB	14	68 GB/s	Gemma 4 E4B	masyadong malaki

Buksan ang anumang Mac para sa buong listahan nito, o magsimula sa isang model o sa memory size mo.

Paano ito kinuwenta

Memory: ang mga parameter ng model sa 8, 6, 4, 3 o 2 bit kada weight dagdag ang group scales ng MLX (kalahating bit kada weight), o ang native format nito, at may espasyo pa para sa context. Bilang default, mga 75% ng unified memory ang ipinapagamit ng macOS sa GPU. Bilis: ang memory bandwidth na hinati sa mga byte na binabasa kada token ang pinakamataas na posible para sa isang usapan; sa mga pampublikong benchmark ng llama.cpp sa Apple Silicon, umaabot ang totoong takbo sa mga 45% hanggang 80% nito, at iyon ang range na ipinapakita. Mga tantiya ito, hindi sukat: sukatin ang sarili mong Mac sa MLX Bench. Ang mga numero ng model ay mula sa opisyal na model card ng bawat isa, at ang sa chip ay mula sa Tech Specs ng Apple. Sinuri noong 2026-10-19. Parehong data ng mlx.app/compare.