Apple and Meta Team Up to Run Muse Spark Open Locally on Four Mac Studios

Apple and Meta have worked together to run Meta's new Muse Spark Open AI model entirely on Mac Studio hardware, with no cloud servers involved. It is an unusual partnership, given how often the two companies have clashed over privacy and advertising. Here, though, both have something to gain.

Muse Spark Open is an open-weight model that Meta announced on October 15. It is built for coding, software engineering and tasks that take several steps to finish. The model is huge, as it has 816 billion parameters, with 42 billion active at any time, and its compressed weights alone take up about 493GB.

You will need more than one Mac Studio

A single Mac cannot hold a model this big. Apple says you need either two M5 Ultra Mac Studios with 512GB of unified memory each, or four with 256GB each. The four-machine setup works today, while the 512GB configuration ships only in late October.

Apple's open-source MLX framework splits the model across the connected Macs, which talk to each other over Thunderbolt 5. In Apple's own tests, four Mac Studios generated more than 36 tokens per second. In demos, the model fixed bugs in an open-source iOS app and cut vertical clips from footage in DaVinci Resolve, all on local hardware.

The catch is the price. The M5 Ultra Mac Studio starts at Rs 6,29,900 in India, so four base units alone cost over Rs 25 lakh before any memory upgrades. This is clearly meant for developers, research labs and companies that do not want sensitive code leaving the office.

What is the Windows equivalent?

On the PC side, the closest match is Nvidia's DGX Station for Windows, which Nvidia built with Microsoft and announced at Computex 2026. It runs on the GB300 Grace Blackwell Ultra chip with up to 748GB of memory, enough to fit a model like Muse Spark Open in a single box instead of four. Asus, Dell, HP, MSI, Gigabyte and Supermicro are expected to start selling these systems this quarter.

It will not be cheap either. The Linux version of the DGX Station is already listed in the US at close to $95,000. So the Mac cluster is likely the cheaper way in, while Nvidia offers a single machine and the CUDA software most AI developers already use. Meta says Muse Spark Open also supports Nvidia, AMD and Intel hardware, but it has not shown the model running on a Windows workstation yet.

For regular PC buyers, a model this size is out of reach on any platform. Laptops and mini PCs with AMD's Ryzen AI Max chips and 128GB of memory can run smaller open models. That includes Meta's own 30-billion-parameter Muse Glimmer, which was released in August and runs on a single GPU.

Muse Spark Open is available on Hugging Face, and developers without the hardware can use it through hosted API providers. Meta's licence allows many uses for free but places limits on some commercial use.