A local release with a substantial hardware setup
Meta has announced the release of Muse Spark Open, an open-weight AI model that the company says can run locally across two Apple M5 Ultra Mac Studios using Apple's MLX framework. The announcement describes a specific desktop configuration, not compatibility with every Mac.
Meta lists 816 billion total parameters, with 42 billion active, and a context capacity of one million tokens. It positions the multimodal model for programming and agent-driven work involving extended sequences of actions. Those specifications describe the model's architecture and intended uses; they do not independently establish how reliably it completes such work.
What Meta says it demonstrated
The company says it tested workflows entirely on local hardware through Muse Code, powered by Muse Spark Open. Its examples include changes to Ice Cubes, an open-source Mastodon application, and a DaVinci Resolve workflow that examined video frames and produced vertical crops.
These examples are Meta's account of its own testing, rather than independent assessments. They illustrate the kinds of workflows the company is targeting, but do not establish that developers will obtain equivalent results on other hardware or with different projects.
Downloadable weights, with licensing conditions
Meta lists Hugging Face among the locations where developers can download the weights. It also says developers can adapt the model using PyTorch, while MLX provides the route for running it on Apple silicon.
The release uses the Muse Model License, which Meta describes as including limited commercial terms that allow many developers to use the model without charge. Open-weight availability should therefore not be read as unrestricted commercial permission.
The practical distinction is between access and deployment: downloadable weights enable local operation and customization, while the described Mac setup still involves two high-end machines. The announcement's significance is that combination of accessible weights and a specified local configuration—not evidence that the model will run across the broader Mac lineup.