Just in
GadgetsiPhone Duo Launch Day: Apple Warns of Limited Walk-In Stock at Apple StoresCryptoFranklin Templeton Bitcoin DRIP ETFs Clear Nasdaq Registration Ahead of LaunchAIMicrosoft AI diagnosis tool MAI-DxO beats doctors on hard cases in Science studyTechRhode Island sues Meta over scam ads on Facebook, Instagram and WhatsAppGadgetsOura sues Ultrahuman again to block Ring Pro smart ring sales in the USTechArc browser shutdown set for December 22: what to do before it stops working
Tech

Meta and Apple team up to run Muse Spark Open, an 816 billion parameter model, on Mac Studios

Meta released Muse Spark Open, a giant open weight AI model tuned with Apple to run on two linked Mac Studios. It is the rivals' first public partnership in years.

USA Publishing Group6 min read
Share
A compact Mac desktop setup on a wooden desk, the kind of Apple hardware Meta tuned its Muse Spark Open model to run on
Photo: Iewek Gnos / Unsplash

Key takeaways

  • Muse Spark Open has 816 billion total parameters, about 42 billion active, and a 1 million token context window.
  • Meta and Apple worked for months to run it through MLX on two 512GB M5 Ultra Mac Studios, which ship in late October.
  • Weights are on Hugging Face and major clouds under a license that allows limited commercial use.

Meta and Apple are working together again, and the reason is an AI model. On October 15, Meta released Muse Spark Open, a huge open weight model that it says runs on a desk setup of linked Mac Studio computers thanks to months of joint work with Apple. For two companies that spent years fighting over privacy and ads, it is a surprising handshake, and it says a lot about where the AI race is heading.

Here is what Meta released, what you need to run it, and why the Apple tie matters.

What Meta released with Muse Spark Open

Muse Spark Open is part of Meta's Muse Spark family. According to MacRumors, it has 816 billion total parameters, with about 42 billion active for each token, and a context window of 1 million tokens. That design is called a sparse mixture of experts, which means only a slice of the model does the work on any given request. Meta has pitched it at coding and agent style tasks.

Meta calls it "the most capable US open weight model to date," as both MacRumors and Engadget reported. Engadget said Meta's own benchmarks show it beating the Chinese models GLM 5.3 and Kimi K3 on agentic tasks and visual reasoning. Those are Meta's numbers, not independent tests, so treat them as a starting point.

The model is not Meta's top system. According to 24/7 Wall St., Meta Chief AI Officer Alexandr Wang said it is not the company's most capable model. Meta's bigger push this month was the launch of Muse Wonder, its first frontier AI model, which stays closed.

Where you can get it

MacRumors reported that the weights can be downloaded from Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. It is also offered through API providers including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. The license allows limited commercial use, according to MacRumors.

Why the Muse Spark Open deal with Apple matters

The headline here is less about the model and more about who helped build the path to run it.

Mark Zuckerberg posted on X that the model "runs great on new Mac Studios thanks to work with @johnternus and Apple," according to 24/7 Wall St. Apple CEO John Ternus replied with congratulations and praised the model's efficiency. Wang said the two teams had worked together for months and that Apple is tuning the model through MLX, its software framework for running AI on Apple chips. He also credited Daniel Gross, a former Apple executive who now works at Meta. MacRumors quoted Wang joking, "We're excited for mommy and daddy to be playing nice again."

That joke lands because the history is rocky. 24/7 Wall St. described this as the first public partnership between the two companies since roughly 2017. Apple's 2021 privacy changes, which let iPhone users block app tracking, hit Meta's ad business hard, and the two have traded shots ever since.

What each side gets

For Meta, Apple hardware gives its open model a credible home outside data centers. Developers who want to run a model privately, on their own machines, now have a supported path.

For Apple, it is a showcase. The company has been under pressure to prove its AI story, and a top open model running smoothly on a Mac Studio makes a strong argument for its chips and unified memory. No financial terms, exclusivity or timeline were disclosed, according to 24/7 Wall St., and neither company said whether more joint work is coming.

What it takes to run it at home

This is not a model for a regular laptop. Engadget reported that it needs about 1TB of memory to run locally.

The hardware

Meta says two Mac Studios with the M5 Ultra chip and 512GB of memory each can hold the full model, according to Engadget and MacRumors. Linked together, the two machines act like a single system with a terabyte of unified memory. MacRumors said each chip offers 1.2 TB/s of memory bandwidth.

There is a catch. The 512GB Mac Studio has not shipped yet. Apple has said only that it arrives in late October, with no price announced, per Engadget. The other option is four Mac Studios with 256GB each, which Engadget put at roughly $38,000. According to 24/7 Wall St., the 256GB M5 Ultra model alone is priced at $10,799.

So the realistic audience is research labs, startups and companies that want to keep data in house, not hobbyists. For a sense of where PC makers are going on local AI, see our look at the Surface Laptop Ultra with Nvidia RTX Spark.

Who is affected

Developers gain a powerful open option they can host themselves or reach through common cloud services. Businesses worried about sending sensitive data to outside AI providers have another choice.

Chinese AI labs are a target too. Engadget noted that Meta has argued the US should not let Chinese labs lead in open source AI, and the benchmark comparisons were aimed squarely at them.

Investors are watching as well. 24/7 Wall St. reported Apple shares slipped 0.28% to $342 on October 15, leaving it with a market value near $4.99 trillion, just under the $5 trillion level it first closed above a day earlier. Meta dipped 0.11% to $736.72. The muted moves suggest Wall Street sees this as a technical partnership, not a business shift, at least for now.

What to watch next

Three things will show whether this is a one off or the start of something bigger.

First, Mac Studio sales and pricing. Once Apple names a price for the 512GB model, the real cost of running Muse Spark Open at home becomes clear.

Second, Meta's third quarter results. Meta has guided for revenue of $61 billion to $64 billion, according to 24/7 Wall St., and analysts will want to hear how open models fit its spending plans. Meta's recent moves, such as its decision to ban TikTok ads on Facebook and Instagram, show it is willing to make sharp strategic calls.

Third, more MLX work. If Apple keeps tuning other open models for its chips, the Mac could become a standard box for local AI. Apple's own hardware calendar is busy too, with a reported touchscreen MacBook Pro on the way.

For more on the companies shaping the industry, visit our tech section.

Frequently asked questions

What is Muse Spark Open?

It is an open weight AI model from Meta with 816 billion total parameters and about 42 billion active per token, aimed at coding and agent tasks.

Is Muse Spark Open free?

The weights are free to download from sites like Hugging Face, and the license allows limited commercial use. Running it requires very expensive hardware.

What Mac do I need to run it?

Meta says two M5 Ultra Mac Studios with 512GB of memory each, or four with 256GB each. The 512GB model is due in late October.

Why are Apple and Meta working together?

Meta wanted a strong local home for its open model, and Apple gets a showcase for its chips. Neither side disclosed financial terms.

Muse Spark Open: Meta and Apple Team Up on Mac AI