Meta and Apple are teaming up to make Meta's latest open-weight AI model, Muse Spark Open, run on Mac computers. The 800-billion-parameter model is optimized for Apple's M5 Ultra-equipped Mac Studio, and Meta calls it the most capable US open-weight model to date. Apple's unified memory architecture allows two 512GB Mac Studios to behave like a single machine with 1TB of memory, which is what the model needs to run locally. That hardware does not come cheap: a four-machine 256GB setup would cost roughly $38,000, putting it out of reach for individual developers but potentially attractive to organizations spending heavily on AI inference tokens.
Both CNBC and Engadget report the same basic deal: Meta is releasing Muse Spark Open today, available on Hugging Face, and it is the first model optimized to run locally on Apple's new Mac Studio hardware. The two outlets differ in emphasis. CNBC frames the partnership in the context of Apple's broader AI strategy, noting prior integrations with OpenAI and Google and investor criticism of Siri's capabilities. Engadget focuses on the technical and cost details, including Meta's claim that the model outperforms leading Chinese models on agentic and visual-reasoning benchmarks.
Despite the collaboration, the practical barrier is steep. Engadget notes that 1TB of RAM is a staggering requirement, even for an open-weight model intended to give developers more control and lower token costs. CNBC adds that Meta's open-source push is part of a wider strategy: the company has released smaller open models before and hopes developers will tinker with this one, then eventually upgrade to more capable proprietary offerings like Muse Wonder. Apple, meanwhile, has been outsourcing cutting-edge model work to partners while it works to improve its own AI features. The sources agree the deal helps both companies, but neither suggests it will immediately put Apple's AI on par with leading chatbots and agents.