Muse Spark Open: Meta's AI Model Now Runs on Apple Macs

Muse Spark Open just landed, and the wildest part is who helped Meta launch it: Apple. On Thursday, Meta released the open-weight AI model and said it runs on desktop Macs, a team-up between two companies that have spent years throwing shade at each other.

According to CNBC, the collaboration lets developers run Meta's biggest open model locally on Apple's high-end Mac Studio computers, with no cloud data center in the loop. Meta CEO Mark Zuckerberg posted that the model "runs great on new Mac Studios thanks to work with @johnternus and Apple."

What Muse Spark Open Actually Is

Muse Spark Open is an open-weight model, meaning anyone can download its parameters and run it on their own hardware. Meta calls it the most capable open-weight model built in the United States. The Verge reports it is natively multimodal and has eight hundred sixteen billion total parameters, tuned for long tasks, software engineering and multi-step reasoning.

CNBC adds that it uses a mixture-of-experts design and ships under a license that allows limited commercial use. Developers can grab it from sites like Hugging Face without extra setup for Apple silicon. Meta also says it will be offered through cloud providers and developer tools, according to Axios.

It is not Meta's strongest model. Meta AI chief Alexandr Wang told CNBC, "These aren't the most capable models that we've developed or released." The company's top systems stay proprietary, and Muse Spark Open is the free-to-download option.

Why Apple Is in the Room

Models this large usually need racks of server GPUs. Wang said many powerful open models are too big for most personal computers, but Mac Studios offer enough memory to make Muse Spark Open practical. Meta says the model runs across two connected Mac Studios with M5 Ultra chips, and CNBC reports setups can link up to four.

Apple also published a technical post about its MLX framework splitting open models across several machines. The pitch is simple: your prompts stay on your desk. Axios notes that because the model runs locally, Meta never sees user prompts or data, which sidesteps the privacy fight that has divided the two companies for years.

From Beef to Besties-ish

The history here is spicy. After Apple rolled out privacy changes in 2021, Meta estimated the move would cut its revenue by ten billion dollars the following year, CNBC reports. Zuckerberg reportedly told staff Facebook needed to "inflict pain" on Apple, according to the Wall Street Journal as cited by Axios.

Now the tone is warm. New Apple CEO John Ternus replied to Zuckerberg, "Congrats to the team! Amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk." Wang was blunter with Axios: "We're excited for mommy and daddy to be playing nice again." He said the two sides had been working together for months, and credited former Apple executive Daniel Gross, now at Meta, with helping bridge the gap.

Neither company would say whether this is the start of something bigger. Axios points out that the real milestone would be Apple putting Meta's models inside Apple Intelligence, which is not happening for now.

The Catch: Your Wallet

Here is the reality check for anyone dreaming of a home AI lab. CNBC says a Mac Studio with the required M5 Ultra chip and two hundred fifty-six gigabytes of memory costs at least ten thousand seven hundred ninety-nine dollars. A version with double that memory arrives later this month, and you need more than one machine.

That puts Muse Spark Open firmly in startup and developer territory for now, not a casual weekend download. Axios adds a safety caveat: open models can have their guardrails stripped by bad actors, which raises worries about cyberattacks and other misuse.

Why the Muse Spark Open Team-Up Matters

The bigger story is where AI is heading. If a model this large can live on a desk, companies get another route that skips both the cloud giants and Nvidia's data center chips. CNBC says the move could signal Apple's growing strength in AI hardware and pose a challenge to OpenAI and Anthropic.

Open models are also a political topic. Washington has debated whether free Chinese systems could undercut American AI, and Axios notes that Nvidia, Reflection AI and Mistral all released new open models in recent days. Meta wants to stay in that race with something developers can tinker with, then scale up.

For Gen Z builders, the takeaway is hopeful but expensive: serious local AI is arriving, just not at student prices yet. For the industry, it is a sign that the AI boom is scrambling old alliances. Keep an eye on tech coverage here as the first benchmarks and real-world tests come in.