Story
Oct 20, 2026
Apple and Meta Make Peace, but Muse Spark’s Mac Bill Starts at $10,799
Apple and Meta have joined forces to run Meta’s open-weight Muse Spark model locally on Mac Studios. The partnership pitches privacy and cheaper cloud-free AI, but the hardware required puts that promise far beyond an ordinary desktop.
For years, Apple and Meta embodied Silicon Valley’s cold war. Apple’s 2021 privacy changes hit Meta’s advertising business, and their leaders later traded public barbs. But the AI race has pushed the rivals toward a pragmatic truce.1
On Thursday, the companies unveiled a collaboration to run Meta’s Muse Spark Open on Apple’s high-end Mac Studios. Meta calls the 816-billion-parameter, open-weight model a tool for coding, video analysis and other agentic work; users can download its weights and run it locally rather than send work to a cloud provider.1 Zuckerberg framed the launch as a hardware-software win, writing that Muse Spark Open “runs great on new Mac Studios” after work with Apple and John Ternus.
2
That local-first pitch has clear appeal: greater control over data, potentially lower cloud bills, and more flexibility than a closed model. A separate account of the release described Muse Spark Open as natively multimodal, built for complex coding and agentic tasks, and available through API platforms.3 Enthusiasts seized on the announcement; Hugging Face chief executive Clément Delangue declared, “Meta is back!”
4
But the fine print makes the breakthrough less democratic than the rhetoric. Apple’s M5 Ultra Mac Studio with 256GB of memory—the minimum configuration cited for the model—costs at least $10,799, and users may need two to four machines to run it. Another report put the memory requirement at at least 1TB, underscoring the scale of the setup.13
The partnership is therefore both a show of confidence in Apple’s chips and Meta’s AI reboot—and a reminder that “local” frontier AI still means buying data-center-grade capacity for the desk. Meta’s AI chief Alexandr Wang says the model is meant to let developers tinker at small scale before expanding; for many, the first hurdle will be affording the experiment.1