Meta launches an open-weight AI model that runs on Macs
TechCrunch Lucas Ropek ● Covered by 18 sources
Meta launched Muse Spark Open, an open-weight AI model built to run on Macs. It’s huge and aimed at coding work, but you’ll need a very powerful Mac to use it locally.
Based on reporting by TechCrunch, Lucas Ropek — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Meta says Muse Spark Open is its “most capable” open-weight model in the United States, and it’s clearly built to make a point. The model has 816 billion parameters and a 1 million token context window, and Meta says its own evaluations put it ahead of other popular open-weight models.
The unusual bit is the Mac support. Muse Spark Open was designed to run locally on Apple computers, but not casually. Engadget reports you’d need at least 1 terabyte of RAM, and Meta says the model should run best on the Mac Studio with Apple’s M5 Ultra chip. That puts it in the territory of serious desktop hardware, not a laptop you toss in a bag.
Meta is pitching the model for coding and agentic work, and it says Muse Spark Open is natively multimodal. The company paired it with Muse Code, its AI coding terminal, and used it on agentic workflows, including fixes for a Mastodon client that involved bug fixes, new features, and a full build.
It also says the model handled tasks inside DaVinci Resolve, including reviewing and cropping footage. That matters because it shows Meta isn’t just talking about benchmark bragging rights. It wants this thing to feel useful on an actual desk, in actual software, if the machine underneath can keep up.
Meta is making Muse Spark Open available now through several API platforms. Open-weight models already have a straightforward appeal: they can be cheaper, more transparent, and less boxed in than closed systems. Meta is leaning hard into that argument, with a model large enough to be impressive and demanding enough to remind everyone that “local AI” still has a price tag.
My take — AI-written commentary, not fact-checked reporting
This is Meta doing what Meta does best: turning open weights into a flex. The pitch is not subtle — freedom, transparency, and a Mac-shaped footnote that says you’ll need absurd hardware to enjoy any of it. Open models keep winning the argument on control; closed models keep winning on not requiring a computer that looks like it should power a small weather station.
Read more about this at: TechCrunch