Can a Mac Studio M3 Ultra, 256GB run local LLMs?

Yes — 44 of the 48 current models on this site fit in its 192 GB of usable memory, and one more if you keep the window shorter than 32k. The strongest it holds, GLM-5.3-Flash, runs at 22 tok/s with a 32k window, worked out from memory bandwidth. Whether that saves you money is a different question: at its $7,099 launch price and 500k tokens a day, it pays back in 2,379 years.

Run the numbers on this machine

What the Mac Studio M3 Ultra, 256GB runs

They do not all hold the same window. The weights are a fixed size, but the key-value cache grows with every token you keep, so what is left of 192 GB after the weights is how far the context goes. On one of the 12 below, this machine's memory is what runs out first: GLM-5.3-Flash stops soonest, at 128k. The rest reach 256k, where the calculator's list ends, or the longest setting their own context limit allows.

Speed and memory are at 32k context, the setting the calculator starts on; the memory column is the weights plus the cache for that much of it. Each speed says how it was arrived at, and every one here is estimated from this machine's 819 GB/s of memory bandwidth rather than taken from a published benchmark. There is a page on how that sum works, and what each size needs, and one on what 256 GB runs, machine by machine. The longest context is the longest setting the calculator offers that this machine still holds the model at, cache included, and each one opens the calculator on that model at that length. A figure tagged memory is one this machine ran out of room for, and the rest are stopped by the model's own limit or by the end of the list.

For scale, the calculator starts from 80 tok/s for a hosted API and times a local machine against it. 2 of the 12 above reach it, and the slowest is 17 tok/s.

32 more fit. Ranked below the twelve above: Qwen3.5 122B-A10B, Gemma 4 31B it, Granite 4.2 30B, GLM-4.7-Flash, Gemma 4 12B, Qwen3.5 9B, Nemotron 3.5 Lightning 30B-A3B, MiniCPM5 2B, Qwen3.5 4B, gpt-oss-120b, Ling 3.0 tiny, Granite 4.2 8B, Mistral Small 4 (119B-2603), Qwen3-Coder 30B-A3B, Qwen3-Coder Next, gpt-oss-20b, Gemma 4 E4B, Devstral 2 123B, LFM2.5 2.6B, Devstral Small 2 24B, Ministral 3 14B, Ministral 3 8B. The other 10 have no intelligence-index score, so the ranking cannot place them: Laguna XS 2.1, LFM3 1.7B, Ornith 1.5 9B, Ornith 1.5 35B-A3B, Ornith 1.5 122B, Spark-X2.5 4B, Spark-X2.6 4B, KAT-Coder V2.5 Dev 35B-A3B, LFM2.5 24B-A2B, Flint 40B-A2.5B. Their pages show what each one needs and what runs it.

The 48 counted above are the current models. This site prices 16 superseded ones as well, because people still run them, and this machine holds 16 of those at 32k, strongest first: Qwen3 235B-A22B Instruct 2507, GLM-4.5-Air, Qwen3 30B-A3B Instruct 2507, DeepSeek-R1-Distill-Qwen-32B, DeepSeek-R1-Distill-Llama-70B, Llama 3.3 70B Instruct, Llama 3.1 8B Instruct at Q4_K_M, Llama 3.1 8B Instruct at Q8_0, Mistral Small 3.2 24B Instruct, Qwen3 32B at Q4_K_M, Qwen3 32B at Q8_0, Qwen3 14B, Llama 4 Scout 17B-16E, Qwen3 8B, Gemma 3 27B it, Gemma 3 12B it. Each of those pages asks whether the model is still worth running.

What has been measured on the Mac Studio M3 Ultra, 256GB

Not one of the speeds above was measured on this machine, and that is not because nobody has run a model on one. This site holds four published speeds taken on this machine, three of them on superseded models the table above leaves out. They are below, as published and again at 32k of context, which is what every other speed on this page is taken at.

A published figure is what its source reports, at the window the model was run at. The 32k figure is that measurement read at 32k of context: every token generated reads the weights and the whole key-value cache, so the longer the window the slower it runs. The figures come from geerlingguy/ai-benchmarks, hostbor.com, macstories.net.

One more, at a shorter window

The table above counts what fits at 32k, the context the calculator starts on. One more model fits this machine at a shorter one. Same weights, smaller cache.

Each window is the longest setting the calculator offers that this machine still holds the model at, and it opens the calculator on that model at that length. The memory figures are the weights plus the key-value cache at that window, against the 192 GB this machine's GPU can use.

Other machines to weigh against the Mac Studio M3 Ultra, 256GB

Every row uses the same defaults as the figures above: 500k tokens a day at 15:1 input to output, 32k context, and each machine's strongest model that fits, counted against the same 48 current models. The GeForce RTX 5090, 32GB is priced as the card alone, so add the PC around it before comparing it with a complete computer. All seven cards here are ranked by what each one holds.

Head to head with the machine that replaced it: Mac Studio M5 Ultra, 256GB. There are 111 machine match-ups on the site.

The specifics

Chip

M3 Ultra — 32-core CPU / 80-core GPU

Memory bandwidth

819 GB/s

Usable by the GPU

192 GB — macOS lets the GPU wire roughly 75% of unified memory by default (llama.cpp discussion #2182: 66.7% at 32 GiB or less, 75% above). Raise it with `sudo sysctl iogpu.wired_limit_mb`. Memory is treated as GB throughout, which is slightly conservative.

Power under load

270 W (published) — Apple's published 'max' figure for this chip (support article 102027), a CPU+GPU stress number; inference typically draws less.

Availability

Discontinued 2026-08-25. Launch price; Apple no longer sells it new, so treat as a refurbished/used reference. 256GB required the 32-core chip.

Sources

Apple Support, Apple Support, Low End Mac