Meta’s latest push into open-weight AI arrived with an unusually specific target: Apple’s Mac ecosystem. The company on Thursday released Muse Spark Open, an 816-billion-parameter model it calls the most capable open-weight model in the United States, designed to run locally on Apple computers.¹

The pitch is substantial. Muse Spark Open has a one-million-token context window, handles multiple modalities and is aimed at complex coding and agentic work. Meta said it paired the system with its Muse Code terminal, using it for tasks including fixes and new features for a Mastodon client, while also testing its ability to review and crop video footage in DaVinci Resolve.¹

But the release’s central tension is written into the specifications. Local operation requires at least 1TB of RAM, according to the launch coverage, and the model is expected to run best on a Mac Studio equipped with Apple’s M5 Ultra chip.¹ That turns a headline about AI running on Macs into a proposition for the most expensive end of the Mac lineup, not a broad desktop breakthrough.

Meta chief executive Mark Zuckerberg cast the rollout as a direct product of the companies’ collaboration, writing that the weights were releasing immediately and that the model “runs great on new Mac Studios” thanks to work with Apple and hardware chief John Ternus.Mark Zuckerberg (finkd)²

For Meta, the Apple-friendly deployment strengthens the case for open-weight models: more flexibility, transparency and potentially lower costs than closed systems. For users, however, openness does not erase the price of entry. Muse Spark Open is available through several API platforms now, but its most ambitious local use remains gated by memory on a scale few desktops can offer.¹