M5 generation
Local LLMs on the M5 Ultra
The M5 Ultra moves 1200 GB/s and takes up to 512 GB of unified memory. That combination runs everything up to DeepSeek V3 at Q4, at roughly 33.3 tokens/sec. Bandwidth sets the speed, memory sets the ceiling, and the tables below give both for every model.
Worth knowing. 1.2 TB/s. The 512 GB tier exists only on the fully specced part, 36 core CPU and 80 core GPU; the configurations most people see stop at 256 GB.
Memory bandwidth
1200 GB/s
Unified memory
96, 256, 512 GB
8B model at Q4
~130 tok/s
Found in: Mac Studio (2026). Specifications verified 2026-09-11against Apple’s published figures.
Mac Studio with the M5 Ultra for local LLMs
The Mac Studio is the desktop that carries the M5 Ultra, and for local inference it is a different class of machine from a Mac mini. The mini stops at the M4 Pro: 64 GB and 273 GB/s. The Studio with the M5 Ultra goes to 512 GB and 1200 GB/s, which is 4.4x the bandwidth and therefore about that much faster on any model that fits both. At 512 GB it holds DeepSeek V3 at Q4_K_M, which no mini can.
Two things to check before ordering a Studio for this: memory is fixed at purchase, so size it for the largest model you expect to run, not the one you run today, and the price gap to a used RTX 3090 or a DGX Spark is on the hardware comparison. The calculator gives the memory and tokens per second for any model on this chip.
What the M5 Ultra runs, at Q4_K_M
Memory figures include a 3 GB reserve for macOS and an 8K context. Speeds come from a model fitted to our own measured benchmarks; the method and its error bars are on the calculator page.
| Model | Needs | 96 GB | 256 GB | 512 GB |
|---|---|---|---|---|
| Llama 3.2 3B | 8 GB | 199 tok/s | 199 tok/s | 199 tok/s |
| Mistral 7B | 16 GB | 138 tok/s | 138 tok/s | 138 tok/s |
| Qwen 2.5 7B | 16 GB | 134 tok/s | 134 tok/s | 134 tok/s |
| Llama 3.1 8B | 16 GB | 130 tok/s | 130 tok/s | 130 tok/s |
| DeepSeek R1 Distill 8B | 16 GB | 130 tok/s | 130 tok/s | 130 tok/s |
| Qwen 2.5 14B | 16 GB | 87 tok/s | 87 tok/s | 87 tok/s |
| Qwen 2.5 32B | 32 GB | 46.5 tok/s | 46.5 tok/s | 46.5 tok/s |
| Llama 3.3 70B | 56 GB | 23.5 tok/s | 23.5 tok/s | 23.5 tok/s |
| Qwen 2.5 72B | 56 GB | 22.9 tok/s | 22.9 tok/s | 22.9 tok/s |
| Mistral Small 4 | 88 GB | 14.4 tok/s | 14.4 tok/s | 14.4 tok/s |
| Mistral Large 2 | 96 GB | 13.9 tok/s | 13.9 tok/s | 13.9 tok/s |
| Mistral Medium 3.5 | 96 GB | 13.4 tok/s | 13.4 tok/s | 13.4 tok/s |
| DeepSeek V4 Flash | 208 GB | — | 6.2 tok/s | 6.2 tok/s |
| Llama 3.1 405B | 288 GB | — | — | 4.4 tok/s |
| DeepSeek V3 | 480 GB | — | — | 33.3 tok/s |
M5 Ultra against the M3 Ultra
Bandwidth went from 819 GB/s to 1200 GB/s, which is 1.47x, and that ratio carries almost directly into generation speed at the same model and quantization. The memory ceiling went from 512 GB to 512 GB, which is the part that decides whether a model runs at all rather than how fast. The ceiling did not move, so this is a speed upgrade rather than a capability one. If a model did not fit before, it still will not.
Common questions
What is the largest model an M5 Ultra can run?
At Q4_K_M with 512 GB of unified memory, DeepSeek V3 is the largest of the major open models that fits, needing about 480 GB once macOS is accounted for. Anything larger has to drop to a smaller quantization, run across clustered machines, or move to a chip with a higher memory ceiling.
How fast is the M5 Ultra for local inference?
It has 1200 GB/s of memory bandwidth, and generation speed on Apple Silicon is set almost entirely by bandwidth divided by the size of the weights. In practice that works out to roughly 130 tokens/sec on an 8B model at Q4 and 46.5 on a 32B, assuming they fit. That is about 1.47x the M3 Ultra at the same model size.
Is a Mac Studio with the M5 Ultra good for local LLMs?
Yes, and it is the reason the Mac Studio exists in this market: it is the only Apple desktop that ships the M5 Ultra with up to 512 GB of unified memory and 1200 GB/s of memory bandwidth. A Mac mini tops out at the M4 Pro with 64 GB and 273 GB/s, so the Studio is the step up when a model does not fit in 64 GB or when you want 4.4x the generation speed on the same model.
How much memory should I order with an M5 Ultra?
Memory decides which models you can run at all, and it cannot be upgraded later, so it is the one specification worth overbuying. M5 Ultra ships in 96, 256, 512 GB. The 96 GB configuration caps you at the small end; 512 GB is what opens up the larger models in the table on this page.
The M5 Ultra is in our lineup, not yet in our racks
Straight answer, since you are probably here deciding what to buy. We do not own this hardware yet. It is listed in our configurator at its real price, $786/mo for the entry configuration, and choosing it puts you on the waitlist rather than through checkout. If you want one, that is the honest way to tell us, and it is what decides which machines we buy next. Available today, the closest thing we run is the M3 Ultra at 819 GB/s, up to 256 GB, from $571/mo, which is 68% of the bandwidth in the table above.