Meta claims its Muse Spark Open model, released today, is the first in its class that runs on a desktop — specifically two Apple M5 Ultra Mac Studios running the latest MLX framework. The company also calls it "the most capable US open-weight model to date." Both quotes come straight from Meta's announcement.

What makes the pairing unusual is who built it. Meta and Apple, typically rivals, worked together on the launch and are now publicly supporting each other on social media. That's a sharp break from years of friction between the two companies, and it signals where each side thinks the local-inference market is heading.

What did the two CEOs actually say?

"Muse Spark Open — weights releasing today — runs great on new Mac Studios thanks to work with @johnternus and Apple," Meta CEO Mark Zuckerberg posted on X. Apple's new CEO John Ternus replied directly: "Congrats to the team! Amazing to run a model of Muse…" on the new hardware.

The public back-and-forth matters because it confirms hands-on engineering collaboration, not just a compatibility footnote. Zuckerberg named Ternus personally; Ternus amplified the launch within hours.

Why M5 Ultra and MLX matter

The technical claim rests on two specifics:

• Hardware: Two M5 Ultra Mac Studios — Apple's top-tier desktop silicon, the configuration shown in Meta's launch materials.
• Software: The latest MLX framework, Apple's machine-learning library optimized for its own silicon.

Meta frames the result as "the first model in its class" that can run at the desktop level on that stack. For developers and researchers who can't book datacenter GPUs, an open-weight frontier-class model that fits on a pair of Mac Studios changes the economics of local experimentation. No cloud bill. No queue.

The commercial logic cuts both ways. Apple gets a flagship AI workload that showcases M5 Ultra's memory bandwidth and compute at a moment when buyers are scrutinizing whether desktop Macs can serve as serious inference boxes. Meta gets its open-weight strategy planted on premium consumer hardware — distribution that its rivalry with Apple never previously allowed.

Rivals, but aligned on local AI

The Wall Street Journal has chronicled the Meta-Apple relationship as a clash of tech titans, and the two companies have sparred over privacy, advertising and platform control for more than a decade. Today's launch doesn't erase that history. It does show both CEOs willing to spend reputational capital on a shared bet: that open-weight models running locally on Apple silicon represent a market worth cooperating on, right now.

The timing is telling. Ternus is newly in the CEO seat, and his first public AI moment is a joint product shout-out with Zuckerberg — a deliberate signal about where Apple plans to compete.

What's next

The weights are available as of today, which means independent benchmarks on real M5 Ultra hardware should follow within days. Those results — not the CEOs' posts — will determine whether two Mac Studios genuinely suffice for this model class, and whether Meta's open-weight push on Apple silicon becomes a template or a one-off.