Apple wants the new MacBook Pro with PROLED XDR Display to be a machine you can run serious AI on without sending a thing to the cloud. In its press release, Apple says the M5 Max version can run large language models with hundreds of billions of parameters entirely on device. The M5 Max model starts at $4,499 for the 14-inch size and $4,799 for the 16-inch, with pre-orders open today and availability on Friday, November 6.
That claim is the biggest AI promise Apple has made for a laptop so far, and it rests on three things: a Neural Accelerator in every GPU core, a large pool of unified memory, and high memory bandwidth. Here’s what Apple’s own release and spec sheet say about each one, and what they leave out.
Apple says M5 Max can run models with hundreds of billions of parameters on the laptop itself
Apple calls M5 Pro and M5 Max its most powerful chips ever built for a pro laptop. It pitches M5 Max at the heaviest jobs: running large language models with hundreds of billions of parameters fully on device, 3D VFX, AI development, film scoring and 8K post-production. M5 Pro gets a slightly different list that still includes AI tasks, alongside code compilation, DNA sequencing and editing multiple streams of 8K ProRes video.
Don’t miss the best of The Mac Observer
Set us as a preferred source and our Apple reporting ranks higher in your Google Search results and Discover feed — one tap, no account changes.
Running a model on device means your prompts, documents and code stay on your Mac. You don’t need a network connection to get an answer, and you don’t pay per request to a cloud provider. Apple doesn’t say which models it tested, how fast they ran, or what quantization it used, so treat “hundreds of billions of parameters” as Apple’s ceiling, not a speed promise.
A Neural Accelerator sits inside each GPU core on both M5 Pro and M5 Max
The headline hardware change for AI is that each GPU core now has its own Neural Accelerator. Apple says this speeds up AI workloads, and the example it gives is running a coding agent on a large language model that lives on the Mac. Both chips also keep a 16-core Neural Engine, according to the spec page.
The GPU counts tell you how many of those accelerators you get. Apple’s specs list them like this:
- 14-inch M5 Pro: 16-core GPU, configurable to 20-core
- 16-inch M5 Pro: 20-core GPU
- 14-inch M5 Max: 32-core GPU, configurable to 40-core
- 16-inch M5 Max: 32-core or 40-core GPU
Because there’s one Neural Accelerator per GPU core, the 40-core M5 Max has the most AI hardware in the lineup. Apple also lists enhanced shader cores, a third-generation ray-tracing engine and Dynamic Caching for the GPU, though it ties those mostly to apps like Blender and Unreal Engine rather than to AI.
Unified memory up to 128GB decides how big a model you can load
For local language models, memory is usually the hard limit. The model’s weights have to fit in memory before it can answer you, and on Apple silicon the CPU, GPU and Neural Accelerators all share the same unified memory pool. Apple says these configurations offer up to 128GB of unified memory, and that the memory is faster than before.
You only get the big options on M5 Max with the 40-core GPU. Here are the memory and bandwidth figures from Apple’s spec page:
- M5 Pro: up to 64GB of memory (48GB maximum on the 14-inch with the 16-core GPU), 307GB/s of memory bandwidth
- M5 Max with 32-core GPU: 36GB of memory, 460GB/s of memory bandwidth
- M5 Max with 40-core GPU: 48GB, 64GB or 128GB of memory, 614GB/s of memory bandwidth
Bandwidth is the other half of the story. When a language model generates text, it reads through its weights again and again, so faster memory generally means faster replies. The 40-core M5 Max doubles the M5 Pro’s bandwidth, going from 307GB/s to 614GB/s.
LM Studio is the app Apple shows running local models on the new MacBook Pro
Apple names LM Studio in the release as an app that can run large language models locally on the new MacBook Pro, and it pictures the app on the PROLED display. LM Studio lets you download open models and chat with them on your own machine. Apple also mentions TouchDesigner, which can use AI to speed up creative work.
Apple gives no benchmark numbers for either app. The release makes qualitative claims about AI and saves its hard figures for things like battery life, which it rates at up to 24 hours based on its own testing in September 2026. Results vary by use and configuration.
Buy the 40-core M5 Max with 128GB if local AI is your main job, otherwise M5 Pro is enough
If you want to run the biggest open models Apple is talking about, you need the M5 Max with the 40-core GPU and 128GB of memory. That’s the only setup that reaches the 128GB ceiling and the full 614GB/s of bandwidth, and it costs well above the $4,499 starting price once you add memory in the configurator.
If you run smaller models for coding help or summaries, the M5 Pro starting at $2,899 still has a Neural Accelerator in every GPU core and up to 64GB of memory. That’s a far cheaper way in. And if you’re waiting for real speed numbers before spending this much, that’s fair. Apple hasn’t published any yet, so independent tests after the November 6 launch will tell you more.
Where to buy the touchscreen MacBook Pro, and the cheaper MacBook Pro deals right now
Apple takes orders for the MacBook Pro with PROLED XDR Display on its online store and in the Apple Store app, and it ships November 6. Amazon doesn’t list it yet. If you don’t need touch, Amazon has the M5 Pro and M5 Max models it sits beside for up to $610 off. See our MacBook Pro deals for every price we checked.
Need a faster charger for 50 percent in 30 minutes? Amazon sells Apple’s 96W USB-C Power Adapter for $79 and the 140W adapter for $99.
MacObserver may earn a commission when you buy through links on this page. It doesn’t change the price you pay.
Discussion