The key facts

  • Muse Spark Open: 816 billion parameters, 1 million token context window, native multimodality
  • Local deployment requires at least 1 TB of RAM; it works best on a Mac Studio with the M5 Ultra chip
  • Meta demonstrated the model paired with Muse Code and in controlling DaVinci Resolve

Meta introduces Muse Spark Open

Meta has released a new open-weights model called Muse Spark Open, which the company calls the 'most capable' open model in the US. Its distinguishing feature is that it was designed from the ground up to work with Mac computers. According to benchmark results published by Meta, the model outperforms other popular open-weight models.

Specifications and requirements

The model uses 816 billion parameters and a 1 million token context window. It is natively multimodal and built for complex programming tasks and agentic work. According to Engadget, running it locally on Apple computers requires at least 1 TB of RAM; the model works best on a Mac Studio with the M5 Ultra chip.

Capability demonstrations

Meta said it tested the model paired with its AI coding terminal Muse Code in various agentic scenarios — including fixing bugs, adding new features, and fully building a Mastodon client. The company also ran the model to work with the video editor DaVinci Resolve: according to Meta, the model could review and frame footage.

Context and availability

Open weights give the industry advantages unavailable with closed proprietary models: greater cost efficiency, transparency, and flexibility. Muse Spark Open is already available through a number of different API platforms.