Meta has released Muse Spark Open, an open-weight model that the company describes as the most capable open-weight model in the United States, TechCrunch reports. Meta published evaluation results on Thursday that it says show better performance than other popular open-weight models. Those results come from Meta and the source does not say they have been independently checked.
What the model is
Muse Spark Open has 816 billion parameters and a 1 million token context window. It is natively multimodal and, according to Meta, designed for complex coding work and agentic tasks, meaning work where the model carries out multi-step jobs rather than answering a single prompt.
Running it on a Mac
The model is designed to run locally on Apple computers, a result of a collaboration between Meta and Apple. The catch is memory: Engadget, cited by TechCrunch, puts the requirement at no less than 1 terabyte of RAM. The model performs best on a Mac Studio with the M5 Ultra chip.
What Meta showed it doing
Meta paired the model with Muse Code, its AI coding terminal, and used it in several agentic workflows. One was work on a Mastodon client, covering bug fixes, new features and a full build. Another was operating DaVinci Resolve, the video editing software, where the model reviewed and cropped footage.
Availability
Meta is offering the model now through a number of API platforms. The source does not name those platforms or give pricing, so anyone planning around it will need to check the individual providers.
Why open weights
TechCrunch notes that open-weight models can be considerably cheaper to run than closed, proprietary ones and offer transparency and flexibility that closed models do not. For a business, that is the main reason to look at a release like this, alongside the demonstrated coding and agent workflows.