Meta releases Muse Spark Open, an 816B open-weight model it says runs on two Mac Studios

Meta released the weights for Muse Spark Open on Oct 15, calling it the most capable US open-weight model to date. It's natively multimodal, with 816B total parameters (42B active) and a 1M-token context window, and Meta says it's built for coding and long-horizon agent work. Meta says it runs locally on two M5 Ultra Mac Studios (up to 512 GB of unified memory each) using Apple's MLX. Weights are on Hugging Face (access request required), Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models; API hosts include Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. The Muse Model License requires a “Built with Muse Spark Open” notice and a separate deal with Meta for large hosted services, so read it before building a product.
How you’d use it: If you want a large open-weight model you can run without the cloud, try Muse Spark Open through OpenRouter first, then pull the weights from Hugging Face for MLX, vLLM or llama.cpp once you know it fits your work.