Log inCreate wallet

MacBook Pro with M3 Pro for local AI.

The M3 Pro reads memory at 150 GB/s, which sets how fast it writes. Memory decides which models fit, and it can’t be upgraded later, so pick the size here.

18 GB20 models run wellGemma 4 12B Unified · 10–18 tok/s36 GB28 models run wellQwen3.6 35B A3B · 40–71 tok/s

By memory size

MemoryRun wellBiggest that runs wellgpt-oss-20b
18 GB150 GB/s20Gemma 4 12B Unified36–63 tok/s
36 GB150 GB/s28Qwen3.6 35B A3B36–63 tok/s

Models: MacBook Pro (14-inch, M3 Pro or M3 Max, Nov 2023), MacBook Pro (16-inch, Nov 2023). Not every memory size is offered with every model; Apple’s Tech Specs list what each one comes with.

Questions

Is the MacBook Pro with M3 Pro good for local LLMs?
With 36 GB it runs 28 of 73 open models well, up to Qwen3.6 35B A3B, and writes about 36–63 tokens a second with gpt-oss-20b. The M3 Pro reads memory at 150 GB/s, which sets how fast it writes.
How much memory should a MacBook Pro with M3 Pro have for AI?
With 18 GB it runs 20 open models well; with 36 GB, 28, up to Qwen3.6 35B A3B. Memory decides which models fit, and it can’t be upgraded later.

Other Macs

The MacBook Pro with other chips:

Head to head:

The M3 Pro for local AI · The best Mac for local LLMs · Every Mac

How it’s worked out

Memory: the model’s parameters at 8, 6, 4, 3 or 2 bits per weight plus MLX’s group scales (half a bit per weight), or its native format, with room for the context on top. By default macOS lets the GPU use about 75% of unified memory. Speed: memory bandwidth divided by the bytes read per token is the ceiling for one conversation; in llama.cpp’s public Apple Silicon benchmarks real runs reach about 45% to 80% of it, which is the range shown. These are estimates, not measurements: measure your own Mac on MLX Bench. Model figures from each official model card, chip figures from Apple’s Tech Specs. Checked 2026-10-19. Same data as mlx.app/compare.

MacBook Pro M3 Pro for local AI: which LLMs it runs, by memory