Meta has introduced Muse Spark Open, described by the company as the most capable open‑weight AI model available in the United States, specifically engineered to operate on Mac computers.
The model utilizes 816 billion parameters and a 1 million token context window, delivering performance that Meta claims surpasses other leading open‑weight models, as evidenced by evaluation results published on Thursday.
Muse Spark Open is a natively multimodal model built for advanced coding and agentic workflows, and it runs natively on Apple hardware. Nonetheless, local deployment demands substantial resources—Meta specifies a minimum of 1 terabyte of RAM. For optimal performance, the model is best executed on the Mac Studio equipped with the M5 Ultra chip, according to Engadget.
When sufficient memory is available, Muse Spark Open can serve as a powerful workhorse for coding and other desktop applications. Meta has paired the model with Muse Code, its AI coding terminal, and integrated it into multiple agentic workflows, successfully implementing bug fixes, new features, and a full build for a Mastodon client.
The company also demonstrated the model’s utility by running it with DaVinci Resolve, a professional video editing suite, where it performed tasks such as reviewing and cropping footage.
Open‑weight models deliver key advantages over closed, proprietary systems, including greater cost efficiency, transparency, and flexibility for developers and researchers.
Meta is currently making Muse Spark Open accessible via multiple API platforms for developers and enterprises.