By memory size
| Memory | Run well | Biggest that runs well | gpt-oss-20b |
|---|---|---|---|
| 16 GB200 GB/s | 22 | Phi-4 Reasoning Vision 15B | 56–100 tok/s |
| 32 GB200 GB/s | 30 | Qwen3.6 35B A3B | 47–84 tok/s |
Models: Mac mini (2023). Not every memory size is offered with every model; Apple’s Tech Specs list what each one comes with.
Questions
- Is the Mac mini with M2 Pro good for local LLMs?
- With 32 GB it runs 30 of 73 open models well, up to Qwen3.6 35B A3B, and writes about 47–84 tokens a second with gpt-oss-20b. The M2 Pro reads memory at 200 GB/s, which sets how fast it writes.
- How much memory should a Mac mini with M2 Pro have for AI?
- With 16 GB it runs 22 open models well; with 32 GB, 30, up to Qwen3.6 35B A3B. Memory decides which models fit, and it can’t be upgraded later.
Other Macs
The Mac mini with other chips:
Other Macs with the M2 Pro:
Head to head:
The M2 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.