Meta announced Muse Spark Open, an open-weight artificial intelligence model it describes as the most capable open model in the United States. The company published evaluation results on Thursday accompanying the launch.
Technical specifications and capabilities
Muse Spark Open uses 816 billion parameters and supports a 1 million token context window. According to Meta, the model is natively multimodal — able to handle multiple input modalities — and is targeted at complex coding tasks and agentic workflows.
Mac compatibility and local runtime requirements
The model was designed to run on Apple computers. However, local operation requires substantial hardware: reports indicate a minimum of 1 terabyte of RAM to run the model locally. Meta notes the model performs best on Mac Studio systems equipped with the M5 Ultra chip.
Demonstrations and use cases
Meta says it paired the model with Muse Code, its AI coding terminal, and deployed it into several agentic workflows. The company demonstrated using Muse Spark Open to produce code fixes, add new features, and perform a full build for a Mastodon client. It also applied the model to video-editing tasks, where it reviewed and cropped footage in DaVinci Resolve.
Advantages of open-weight models and availability
Meta highlighted industry advantages of open-weight models compared with closed, proprietary systems, including cost efficiency, transparency, and flexibility. Muse Spark Open is available now through multiple API platforms.
Practical considerations
While the model’s Mac support and capabilities are notable, the large memory requirements mean many users will likely use the model via API rather than running it locally unless they have high-end hardware investments. Local deployment carries significant hardware cost and complexity.
