Meta Releases Muse Spark Open, an 816B Open-Weight Model Built With Apple to Run on Mac Studios
Meta has released the weights for Muse Spark Open, an 816-billion-parameter mixture-of-experts model it calls the most capable US open-weight model to date, and worked with Apple so it can run locally across two to four M5 Ultra Mac Studios.
By 1EyeAI Desk ·
Meta Superintelligence Labs released the weights for Muse Spark Open on Oct. 15, 2026: a natively multimodal mixture-of-experts model with 816 billion total and 42 billion active parameters and a 1-million-token context window. Meta says it is built for coding and agentic work and calls it “the most capable US open-weight model to date.” That is a company claim; independent benchmarks are not in yet.
The weights are on Hugging Face, alongside a 4-bit NVFP4 version, and the model is offered through hosted API partners including Together AI, Fireworks AI, Groq and OpenRouter. It ships under the Muse Model License, which Meta describes as having limited commercial terms. It is less capable than Meta’s proprietary Muse Wonder, released earlier this week.
Apple published a technical post the same day showing the model running on its open-source MLX framework, split across several Mac Studios linked by Thunderbolt 5. Apple says the 4-bit model needs about 493GB for its weights alone, too much for one machine, and reports roughly 15.7 tokens per second on two M5 Ultra Mac Studios and about 36.7 on four, at an 8,192-token context.
It is not cheap. CNBC reports that a Mac Studio with M5 Ultra and 256GB of memory starts at $10,799, with a 512GB version due later this month, and you need at least two. Meta says the model met the release thresholds of its Superintelligence Scaling Framework, with a full evaluation report “available soon.”
What this means for you: Most people will not run this at home, but developers can try it today through hosted APIs or Hugging Face. If you have serious Mac hardware, it is a real option for private, local coding agents. Test it on your own tasks before trusting Meta’s benchmark claims.
Sources
- Meta AI Research: Introducing Muse Spark Open
- Apple Machine Learning Research: Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio
- Hugging Face: Muse Spark Open collection
- CNBC: Apple and Meta team up to make Muse AI model run on Mac computers
- 9to5Mac: Meta launches new Muse Spark Open model, optimized to run on Mac Studio