Meta releases the weights for Muse Spark Open, an 816-billion-parameter model built to run on two Mac Studios

Meta released Muse Spark Open on October 15 and calls it the most capable open-weight model from a US company to date. The model is natively multimodal, has 816 billion total parameters with 42 billion active, and has a context window of 1 million tokens. Meta says it is built for coding, long-horizon agentic tasks, and multi-step reasoning. Meta says it is the first model in its class that can run on a desktop, using two Apple M5 Ultra Mac Studios with Apple's MLX framework. TechCrunch, citing Engadget, reports that running it locally requires at least 1 terabyte of RAM. The Verge reports that the CEOs of Meta and Apple have both publicly backed the companies' work on the model. The model is available through API providers including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. The weights can be downloaded from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models. Meta says the model runs on inference engines including SGLang, vLLM, TensorRT-LLM, llama.cpp and Apple MLX, with optimizations for AMD, Apple, Intel and NVIDIA chips. According to TechCrunch, Meta paired the model with its Muse Code terminal. Meta used it to fix bugs and add features in a Mastodon client and to review and crop footage in DaVinci Resolve. Meta's claims that the model outperforms other open-weight models come from its own evaluations.