M3 Max (GPU count not published, 64 GB) — benchmark record

M3 Max (GPU count not published, 64 GB) local LLM benchmarks on Apple Silicon Mac with 24 published rows across 2 models. Peak published speed is 10.4 tok/s on Qwen 3 32B. Published runtimes include llama.cpp, Ollama.

24Benchmark rows

2Models tested

10.4Fastest avg tok/s (Qwen 3 32B)

0Lab benchmarks

Quick take

Best published speed here is 10.4 tok/s on Qwen 3 32B at Q8_0. Longest published context on this page is 32k. This page is evidence, not the full buying answer.

Based on 24 external benchmarks; no lab runs yet.

Published runtimes: llama.cpp, Ollama.

Need a recommendation? Use Buy Checking a target model? Use Fit Want a setup answer? Use Run Browse Macs by exact hardware Need the full audit trail? Use Bench

Current published coverage

Published model coverage includes Qwen 3 32B, Llama 3.3 70B. Published runtime coverage on this chip includes llama.cpp, Ollama. Fastest published row is 10.4 tok/s on Qwen 3 32B at Q8_0. Longest published context is 32k.

Qwen 3 32B Llama 3.3 70B

Raw benchmark rows for M3 Max (GPU count not published, 64 GB)

Rows sorted by avg tok/s descending. Click source badge to see original measurement page.

Model	Quant	RAM req.	Context	Avg tok/s	Prompt tok/s	Runtime	Source
Qwen 3 32B	Q8_0	—	264	10.4 tok/s	153.6 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	264	10.3 tok/s	152.1 tok/s	Ollama	ref
Qwen 3 32B	Q8_0	—	450	10.3 tok/s	169.5 tok/s	Ollama	ref
Qwen 3 32B	Q8_0	—	723	10.3 tok/s	164.8 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	450	10.3 tok/s	171.4 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	723	10.3 tok/s	163.8 tok/s	Ollama	ref
Qwen 3 32B	Q8_0	—	1k	10.2 tok/s	169.2 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	1k	10.1 tok/s	168.3 tok/s	Ollama	ref
Qwen 3 32B	Q8_0	—	2k	10.1 tok/s	167.0 tok/s	Ollama	ref
Qwen 3 32B	Q8_0	—	2k	10.1 tok/s	166.8 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	3k	9.9 tok/s	162.2 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	3k	9.9 tok/s	161.5 tok/s	Ollama	ref
Qwen 3 32B	Q8_0	—	5k	9.7 tok/s	154.2 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	5k	9.7 tok/s	153.0 tok/s	Ollama	ref
Qwen 3 32B	Q8_0	—	8k	9.2 tok/s	140.1 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	8k	9.2 tok/s	139.0 tok/s	Ollama	ref
Qwen 3 32B	Q8_0	—	12k	8.6 tok/s	128.0 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	12k	8.6 tok/s	127.1 tok/s	Ollama	ref
Llama 3.3 70B	Q4_K - Medium	—	258	8.2 tok/s	67.9 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	20k	7.6 tok/s	111.2 tok/s	llama.cpp	ref
Qwen 3 32B	Q8_0	—	20k	7.5 tok/s	111.8 tok/s	Ollama	ref
Llama 3.3 70B	Q4_K - Medium	—	8k	7.5 tok/s	65.2 tok/s	llama.cpp	ref
Llama 3.3 70B	Q4_K - Medium	—	16k	7.0 tok/s	59.5 tok/s	llama.cpp	ref
Llama 3.3 70B	Q4_K - Medium	—	32k	6.1 tok/s	50.3 tok/s	llama.cpp	ref

Models tested on this chip

Qwen 3 32B Llama 3.3 70B

Next step

Use Rankings when you need the best overall answer for a Mac, and keep this page open when you need the evidence behind that recommendation.

If you are comparing local Apple Silicon against rented cloud hardware, use AI Data Center Index for current GPU rental context.

Data

benchmarks.json — full dataset · chips.json — chip summaries · benchmarks.csv — CSV export

Data from in-house lab measurements plus community-published benchmarks. See all chip families →