Meta released the weights for Muse Spark Open on October 15, calling it the most capable US open-weight model to date. The natively multimodal model has 816 billion total parameters with 42 billion active, a 1-million-token context window, and is built for coding and agentic work, including long-horizon tasks and multi-step reasoning.
The release was optimized with Apple for desktop use: Meta says Muse Spark Open is the first model in its class to run locally on two Apple M5 Ultra Mac Studios with the latest MLX framework. Apple's unified memory architecture gives the M5 Ultra's integrated GPU up to 512 GB per machine, so two Studios hold the full model, and 1.2 TB/s of memory bandwidth keeps token reads fast. Meta describes the pair as behaving like one supercomputer with a terabyte of unified memory.
Apple CEO John Ternus congratulated Meta's team publicly, saying it was "amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk." MacRumors notes Apple has not yet released a 512 GB Mac Studio; it is expected later this month, and the current 256 GB model already sells for more than $10,000, so the two-machine setup Meta describes carries a five-figure hardware cost.
Beyond Apple silicon, Muse Spark Open is available through API platforms Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Weights are downloadable from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models. It runs on SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp and Unsloth, with optimizations for AMD, Apple, Intel and NVIDIA silicon. The release ships with custom auto-tunable Helion GPU kernels in SGLang, which Meta says deliver 1.03 to 3.95 times kernel-level speedups on NVIDIA and AMD GPUs.
Meta says it tested agentic workflows running entirely on-device with Muse Code powered by Muse Spark Open, including shipping real changes to Ice Cubes, an open-source Mastodon client, and driving DaVinci Resolve end to end. Those demonstrations are Meta's own; independent benchmarks were not part of the release materials.
The license allows limited commercial use, according to MacRumors; teams considering production deployment should read the license terms rather than assume full open-source freedoms.