M1 Max LLM benchmark
No signed M1 Max submissions yet. Check the configuration guidance below, or contribute a reproducible measurement from your own machine.
Specs vendor specs, not measured
- Unified memory (max configuration)
- 64 GB
- Memory bandwidth
- 400 GB/s
- Launch date
- 2021-10-18
Sources: support.apple.com, apple.com. Vendor-published figures, accessed 2026-10-14.
Plan memory before choosing a model
The catalog lists a family maximum of 64 GB of unified memory. Enter your installed amount; smaller configurations have less space. Model weights, context cache and runtime overhead must share the available memory. Capacity alone does not establish which model will fit or how fast it will run. Use the memory-fit calculator for your model and context as an estimate, then verify with the intended runtime. No signed M1 Max measurement is available yet.
Contribute a signed M1 Max benchmark
Run on YOUR hardware to populate this page: pipx install llm-speed && llm-speed bench
$ pipx install llm-speed && llm-speed bench
Community folklore on M1 Max
5 unverified claims extracted from Reddit/HN comments. Lower trust than signed runs above; every row links to the source.
- communityconfidence 75%
10.62tok/s — qwen2.5-32b on M1 Max via ollama q4_K_M
our signed data: M1 Max · qwen2.5-32b
“is the sky blue?’ runs like this (ollama - m1 max 2e/8p/32 gpu): qwen2.5-14b-instruct-q4_K_M - 26.55 tokens/s qwen2.5-32b-instruct-q4_K_M - 10.62 tokens/s”
- communityconfidence 75%
26.55tok/s — qwen2.5-32b on M1 Max via ollama q4_K_M
our signed data: M1 Max · qwen2.5-32b
“f you tell me the prompt you’re using. ‘why is the sky blue?’ runs like this (ollama - m1 max 2e/8p/32 gpu): qwen2.5-14b-instruct-q4_K_M - 26.55 tokens/s qwen2.5-32b-instruct-q4_K_M - 10.62 tokens/s”
- communityconfidence 60%
11.84tok/s — Qwen2.5-32B on M1 Max via mlx
our signed data: M1 Max · Qwen2.5-32B
“This is great information! You're getting roughly double the speed as my M1 Max 32-core GPU 64GB. I got 11.84 tokens/s on Qwen2.5-32B-Instruct (4bit) MLX”
- communityconfidence 60%
16.00tok/s — Gemma3 27B on M1 Max via mlx
our signed data: M1 Max · Gemma3 27B
“My friend just got MacBook Pro M1 Max 64GB for $1200 used. **Gemma3 27B Q4** on MLX does 16tok/s on that. 800GB/s memory bandwidth. Maybe consider that?”
- communityconfidence 50%
57.00tok/s — on M1 Max via lm-studio
our signed data: M1 Max
“Max (64GB, 24 GPU)|LM Studio|**17.0** (56.6)|**13.4** (56.8)|**5.9** (54.4)|**38.3** (58.9)| Generation speed is virtually identical (\~54-57 tok/s both). The difference is entirely in prefill: oMLX is up to **10x faster** on long contexts. At 8K context (prefill-test turn 4), L…”
Request a run on M1 Max
No measured run yet for these pairs. Vote for the one you want; the most-requested runs go first. See all requests.
We never sell or share your email. Privacy