Mark Zuckerberg and Apple just became collaborators. Let that sink in for a second.

On Thursday, Oct. 15, Meta released Muse Spark Open, an 816-billion-parameter open-weight AI model, and announced that Apple had worked with it for months to make the model run locally on Mac Studio computers. It’s the first public technical partnership between the two companies in years, and it hands developers a way to run frontier-scale AI on hardware sitting on their desks instead of in someone else’s data center.

The Wall Street Journal called it what it is: a new chapter for two companies that have long been foes. And the numbers explain why this one matters more than a typical product announcement.

What Muse Spark Open actually is

Muse Spark Open is big, even by 2026 standards. It carries 816 billion total parameters, though only 42 billion are active during processing, thanks to a mixture-of-experts architecture that routes each request to the relevant slice of the model. Meta built it for coding and agentic work, meaning long-horizon tasks and multi-step reasoning, the kind of software engineering that used to require a closed frontier model. It handles text, image, video and audio natively, and it supports a 1 million token context window.

Meta describes it as the most capable open-weight model built in the U.S. to date. Its weights are free to download, and developers can run or modify the model under a license that permits commercial use. Meta also made it available through cloud providers and developer tools. But the headline feature, literally and figuratively, is the Apple part.

“We’re excited for mommy and daddy to be playing nice again,” Meta’s AI chief Alexandr Wang told Axios, according to the outlet’s report. Wang said the companies had worked together for several months, and that Daniel Gross, a former Apple executive now at Meta, played a key role in bringing the two sides together.

How it runs on a desk

Here’s the engineering part. Even compressed, Muse Spark Open’s weights take up about 493GB, according to Apple’s own research. No single consumer machine holds that. Apple’s answer was to update its open-source MLX framework so the model can be split across multiple Mac Studios.

The setup: pair two Mac Studios with 512GB of unified memory each, or four machines with 256GB each, and they share the workload. Apple’s M5 Ultra chip provides up to 512GB of unified memory with 1.2TB/s of memory bandwidth, and the machines talk to each other over Thunderbolt 5 using remote direct memory access, which keeps the CPU out of the data-shuffling path. In Apple’s testing, four Mac Studios generated more than 36 tokens of text per second.

That speed doesn’t come cheap. A single 256GB Mac Studio starts at $10,799, and a four-machine configuration can run past $40,000, CNBC reports. Apple’s 512GB Mac Studio option isn’t expected to ship until late October. So yes, frontier-scale AI now fits on a desk. It just costs about as much as a car to get there.

Why this is a surprise

Apple and Meta’s feud is one of the longest-running in Silicon Valley. Apple executive chairman Tim Cook and Zuckerberg have clashed repeatedly over privacy and business models. In 2018, during Facebook’s Cambridge Analytica scandal, Cook was asked what he would do if he were Zuckerberg. “I wouldn’t be in this situation,” he replied. Zuckerberg called the comment “extremely glib.” Then in 2021, Apple introduced privacy changes that kneecapped Meta’s ad targeting, and Zuckerberg reportedly told employees that Meta needed to “inflict pain” on Apple, according to the Journal. Meta estimated in early 2022 that the change would cut its revenue that year by $10 billion.

The chemistry on Thursday was different. Zuckerberg posted on X that “Muse Spark Open — weights releasing today — runs great on new Mac Studios thanks to work with @johnternus and Apple.” Apple CEO John Ternus replied: “Congrats to the team! Amazing to run a model of Muse Spark Open’s caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI!”

Representatives from both companies declined to say whether this is the start of a deeper relationship, according to CNBC.

What both sides get

For Meta, the deal is a distribution story. This is only the second open-weight model Meta has released since it overhauled its AI efforts last year, and Wang told the Journal the company believes much of the world will run on open models, and “they should be American.” The timing is pointed: Nvidia released three new open-weight Nemotron models the same day, and Meta acknowledged that Chinese open models from companies like Alibaba and DeepSeek are still ahead for software development and some other tasks. Pairing with Apple gives Meta a showcase for its model that no data-center benchmark can match.

For Apple, it’s a shot of credibility in the one arena where the company keeps getting graded as a laggard. Mac Studio and Mac Mini computers have already become popular among AI developers who want to run models locally, thanks to Apple’s unified memory architecture. Working directly with Meta on distributed inference turns that enthusiast habit into an official, supported workflow. It’s also a way to put a flagship AI story on Apple hardware without building the model itself.

And for developers, the pitch is practical: running the model locally means Meta never sees your prompts or your code. No cloud bills, no data leaving the building.

Why it matters

This announcement is really about two shifts happening at once. First, open-weight models have gone from scrappy alternatives to the center of the AI race, and both companies just admitted that the battlefield matters enough to team up on it. Second, the rivalry map in Silicon Valley is being redrawn by AI faster than anyone expected. Apple and Meta avoided public partnerships for years. A shared interest in bringing serious AI to personal computers just erased that rule.

Sources: The Wall Street Journal, CNBC, Axios, AppleInsider, The Tech Portal

FAQ

What is Meta Muse Spark Open? Muse Spark Open is Meta’s 816-billion-parameter open-weight AI model, released Oct. 15, 2026. It uses a mixture-of-experts design with 42 billion parameters active at a time, supports a 1 million token context window, handles text, image, video and audio, and is built for coding and agentic work. Meta describes it as its most capable open-weight model built in the U.S. to date.

Can Muse Spark Open run on a Mac? Yes. Meta and Apple jointly announced that the model runs locally across two to four Mac Studio computers with Apple’s M5 Ultra chips, using Apple’s MLX framework for distributed inference. It needs two machines with 512GB of unified memory each, or four machines with 256GB each, because the model’s compressed weights alone occupy about 493GB.

Why are Apple and Meta working together on AI? Both companies want developers to have a powerful American-made open-weight model they can run on personal hardware instead of rented cloud servers. Meta’s AI chief Alexandr Wang told Axios the two teams worked together for several months, with former Apple executive Daniel Gross playing a key role. The partnership is striking because Apple and Meta have been public rivals for years.

How much does it cost to run Muse Spark Open on Macs? A single 256GB Mac Studio starts at $10,799, and a four-machine setup can run past $40,000, according to CNBC. Apple’s 512GB Mac Studio option is expected to ship in late October 2026. The model’s weights are free to download under Meta’s license, which permits commercial use.