Is the Mac Studio with M3 Ultra good for local LLMs?

With 512 GB it runs 60 of 73 open models well, up to DeepSeek V4.1 Flash 1017, and writes about 194–345 tokens a second with gpt-oss-20b. The M3 Ultra reads memory at 819 GB/s, which sets how fast it writes.

How much memory should a Mac Studio with M3 Ultra have for AI?

With 96 GB it runs 43 open models well; with 512 GB, 60, up to DeepSeek V4.1 Flash 1017. Memory decides which models fit, and it can’t be upgraded later.

Mac Studio M3 Ultra for local AI: which LLMs it runs, by memory

By memory size

Memory	Run well	Biggest that runs well	gpt-oss-20b
96 GB819 GB/s	43	gpt-oss-120b	194–345 tok/s
256 GB819 GB/s	53	MiMo V2.6 Flash	194–345 tok/s
512 GB819 GB/s	60	DeepSeek V4.1 Flash 1017	194–345 tok/s

Models: Mac Studio (2025). Not every memory size is offered with every model; Apple’s Tech Specs list what each one comes with.

Questions

Is the Mac Studio with M3 Ultra good for local LLMs?: With 512 GB it runs 60 of 73 open models well, up to DeepSeek V4.1 Flash 1017, and writes about 194–345 tokens a second with gpt-oss-20b. The M3 Ultra reads memory at 819 GB/s, which sets how fast it writes.
How much memory should a Mac Studio with M3 Ultra have for AI?: With 96 GB it runs 43 open models well; with 512 GB, 60, up to DeepSeek V4.1 Flash 1017. Memory decides which models fit, and it can’t be upgraded later.

Other Macs

The Mac Studio with other chips:

Head to head:

Against PCs and graphics cards:

M3 Ultra vs DGX Spark

The M3 Ultra 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.