Meta releases an open-weight AI model that can run on Mac computers

Meta has introduced Muse Spark Open, calling it the most capable open-weight model in the United States. The model is built around 816 billion parameters and a 1 million token context window. Meta says evaluation results released on Thursday present Muse Spark Open as superior to other popular open-weight models, while its Mac-focused local execution design makes the release notable beyond the evaluation claims.

Background: An Open-Weight Model Aimed at Mac Workflows

Muse Spark Open is natively multimodal and built for complex coding tasks and agentic labor. Its distinctive design point is that it has been made to work with Macs, moving the conversation from hosted model access toward local desktop use. Meta frames the broader open-weight advantage in practical terms: open-weight models can be significantly more cost-efficient and can provide transparency and flexibility that closed and proprietary models do not allow.

The Mac angle matters because it places the model inside a local desktop path rather than only a remote API path. Meta’s stated local execution goal makes memory a central constraint for Mac use.

Development: 816 Billion Parameters, Long Context and Agentic Coding

The model’s 816 billion parameters and 1 million token context window are central to Meta’s positioning. In released evaluations, the company claims stronger performance than other popular open-weight models. Meta also says it paired Muse Spark Open with Muse Code, its AI coding terminal, and deployed the model across various agentic workflows.

Two examples illustrate the intended workload. For a Mastodon client, Meta says the model supported bug fixes, new features and a full build. For video editing, Meta deployed it in the service of running DaVinci Resolve, the high-end video editing software, and said the model was capable of reviewing and cropping footage.

These examples show Meta presenting Muse Spark Open as suited to sustained coding and application-control tasks. The combination of multimodal capability, coding terminal pairing and agentic workflow deployment is presented as the practical value proposition.

Latest Progress: Local Use Requires High Memory

Despite the Mac design, local execution has a steep memory requirement. Engadget reports that running the model locally needs at least 1 terabyte of RAM. Meta says the model will run best via the Mac Studio equipped with the M5 Ultra chip. The company is making Muse Spark Open available now through a number of different API platforms.

The API platforms provide a route for users who do not have the memory needed for local deployment. The Mac Studio M5 Ultra guidance, meanwhile, identifies the hardware path for the most capable local Mac use described by Meta.

Current Stage: A High-Capability Model Behind a High-Memory Gate

Muse Spark Open now sits at the intersection of open-weight ambition and desktop hardware limits. Its 816 billion parameters, 1 million token context window, multimodal design and agentic coding examples show the use cases Meta is emphasizing. The next phase will be defined by how developers use it through API platforms and whether Mac users can work within the 1 terabyte RAM requirement and the Mac Studio M5 Ultra guidance.

Meta description: Meta launches Muse Spark Open, an 816 billion parameter open-weight model with a 1 million token context window, designed for Macs and coding tasks.

Tags: Meta AI, Muse Spark Open, open-weight models, Mac Studio, agentic coding

Featured image: Abstract blue and purple neural network over a computer chip, no people, no logos, no text.