Meta Introduces Open-Weight AI Model Compatible with Mac Computers

Meta has introduced what it describes as the most advanced open-weight model in the United States, named Muse Spark Open. This model has been specifically designed to operate efficiently on Mac computers.

With an impressive architecture encompassing 816 billion parameters and a context window of 1 million tokens, Muse Spark Open aims to outperform existing popular open-weight models, as highlighted by the evaluation results released by Meta on Thursday.

Featuring a native multimodal setup, Muse Spark Open is engineered for intricate coding tasks and various agentic workflows. However, running this model locally demands substantial computing power, necessitating a minimum of 1 terabyte of RAM. It is best suited for systems like the Mac Studio, which is powered by the M5 Ultra chip.

For those capable of meeting the memory requirements, Muse Spark Open is positioned to serve as a robust tool for coding and other desktop applications. Meta has integrated this model with Muse Code, its AI coding terminal, employing it for a range of workflows, including enhancements and bug fixes for a Mastodon client that required comprehensive build updates.

In addition to its coding prowess, Muse Spark Open has also been employed in the operation of high-end video editing software DaVinci Resolve, demonstrating its ability to review and crop footage effectively.

Open-weight models present numerous advantages over their proprietary counterparts, including cost efficiency and greater transparency and flexibility—features often lacking in closed-source models.

Meta is making Muse Spark Open accessible through various API platforms as of now.