M5 generation

Local LLMs on the M5 Pro

The M5 Pro moves 307 GB/s and takes up to 64 GB of unified memory. That combination runs everything up to Qwen 2.5 72B at Q4, at roughly 6.2 tokens/sec. Bandwidth sets the speed, memory sets the ceiling, and the tables below give both for every model.

Memory bandwidth

307 GB/s

Unified memory

24, 48, 64 GB

8B model at Q4

~47.3 tok/s

Found in: MacBook Pro 14/16-inch, Mac mini (2026). Specifications verified 2026-09-11against Apple’s published figures.

What the M5 Pro 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.

ModelNeeds24 GB48 GB64 GB
Llama 3.2 3B8 GB93 tok/s93 tok/s93 tok/s
Mistral 7B16 GB51 tok/s51 tok/s51 tok/s
Qwen 2.5 7B16 GB49.3 tok/s49.3 tok/s49.3 tok/s
Llama 3.1 8B16 GB47.3 tok/s47.3 tok/s47.3 tok/s
DeepSeek R1 Distill 8B16 GB47.3 tok/s47.3 tok/s47.3 tok/s
Qwen 2.5 14B16 GB27.9 tok/s27.9 tok/s27.9 tok/s
Qwen 2.5 32B32 GBtight13.4 tok/s13.4 tok/s
Llama 3.3 70B56 GB6.4 tok/s
Qwen 2.5 72B56 GB6.2 tok/s
Mistral Large 296 GB
Llama 3.1 405B288 GB
DeepSeek V3480 GB

What it cannot run

Even at 64 GB, these need more unified memory than the M5 Pro can address at Q4: Mistral Large 2, Llama 3.1 405B, DeepSeek V3. Dropping the quantization buys a little room but not a generation of it, so the real options are a chip with a higher ceiling or several machines with pooled memory.

M5 Pro against the M4 Pro

Bandwidth went from 273 GB/s to 307 GB/s, which is 1.12x, and that ratio carries almost directly into generation speed at the same model and quantization. The memory ceiling went from 64 GB to 64 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 Pro can run?

At Q4_K_M with 64 GB of unified memory, Qwen 2.5 72B is the largest of the major open models that fits, needing about 56 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 Pro for local inference?

It has 307 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 47.3 tokens/sec on an 8B model at Q4 and 13.4 on a 32B, assuming they fit. That is about 1.12x the M4 Pro at the same model size.

How much memory should I order with an M5 Pro?

Memory decides which models you can run at all, and it cannot be upgraded later, so it is the one specification worth overbuying. M5 Pro ships in 24, 48, 64 GB. The 24 GB configuration caps you at the small end; 64 GB is what opens up the larger models in the table on this page.

The M5 Pro 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, $243/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 M4 Pro at 273 GB/s, up to 64 GB, from $200/mo, which is 89% of the bandwidth in the table above.

Other Apple Silicon