Meta Releases Muse Spark Open, an 816B-Parameter Open-Weight Model Built to Run on Apple's Mac Studio

META on October 15 released Muse Spark Open, which it calls the "most capable U.S. open-weight model to date". The sparse mixture-of-experts model has 816 billion total parameters, about 42 billion of them active per token, and a 1 million-token context window, and Meta is pitching it at coding and agentic work such as multi-step reasoning and software engineering.

The launch doubles as an unusual collaboration with AAPL. Meta tuned the model so it can run locally on two Mac Studios with 512GB M5 Ultra chips using Apple's MLX framework, and says the chip's 1.2 TB/s of memory bandwidth lets the pair behave like one machine with a terabyte of unified memory. Apple CEO John Ternus publicly congratulated the team, and Meta AI chief Alexandr Wang told Axios he was "excited for mommy and daddy to be playing nice again." The 512GB Mac Studio is not on sale yet and is expected later this month, while the 256GB version already costs over $10,000; Engadget estimates a four-machine 256GB setup at about $38,000.

The model is not tied to Apple hardware. It is available through API hosts including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI, with weights on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models. It is free to download under a Muse Model License with limited commercial terms; Unite.AI reports that companies with more than $20 million in revenue that want to offer it as a service must first sign a separate agreement with Meta.

Meta's own benchmark table, as reported by Unite.AI, shows the model behind the leading Chinese open-weight models on headline measures: it scores 39 on the Artificial Analysis Intelligence Index against 45 for GLM 5.3, 44 for Kimi K3 and 40 for Qwen3.8, and 89.0 on GPQA Diamond against 93.5 for Kimi K3 and Qwen3.8. Meta says a full evaluation report is coming. Meta shares were little changed, at $736.72, down about 0.1% according to SentiSense price data. The things to watch are independent benchmark results, developer uptake on the API hosts, and Apple's pricing for the 512GB Mac Studio.