Meta has launched Muse Spark Open, which it calls the most capable open weight model in the United States, TechCrunch reported. The model carries 816 billion parameters and a 1 million token context window, is natively multimodal, and was built for complex coding tasks and agentic work, according to evaluation results Meta released on Thursday.
The local computing angle is the release’s signature. Muse Spark Open was designed to run on Apple’s computers, although the hardware bar is high: at least 1 terabyte of RAM, with the model running best on a Mac Studio equipped with the M5 Ultra chip, according to reporting cited by TechCrunch. Meta said it paired the model with Muse Code, its AI coding terminal, and used it on agentic workflows including fixes for a Mastodon client and reviewing and cropping footage in DaVinci Resolve.
The release is distinct from the Muse Wonder frontier models covered here on October 13. Wonder is Meta’s closed frontier line sold through its API. Spark Open is the open weight counterpart, available now through a number of API platforms as well as for local deployment, and it extends the strategy Meta has used throughout the AI race: give the weights away, own the ecosystem that forms around them, and let openness do the recruiting that sales teams cannot.