Meta has unveiled what it describes as the "most capable" open-weight model to emerge from the United States. Named Muse Spark Open, this model stands out for being specifically engineered to operate on Macs.

The newly introduced model harnesses 816 billion parameters alongside a 1 million token context window, and according to evaluation results Meta published on Thursday, it claims to outperform other widely used open-weight models. Muse Spark Open is natively multimodal and tailored for intricate coding work and agentic tasks. It has likewise been engineered to run locally on Apple's machines.

That said, running the model locally demands substantial hardware, requiring at least 1 terabyte of RAM, per Engadget. Optimal performance comes through the Mac Studio outfitted with the M5 Ultra chip. Even so, provided you can supply the necessary memory, Muse Spark Open could serve as a formidable workhorse for coding and other desktop operations.

Meta reports that it combined the model with Muse Code, its AI coding terminal, and put it to work across a range of agentic workflows, among them coding fixes for a Mastodon client that encompassed "bug fixes, new features, [and] full build."

The company also says it deployed the model to operate DaVinci Resolve, the high-end video editing software, noting that the model proved capable of reviewing and cropping footage.

Open-weight models deliver a set of distinct industry benefits that closed and proprietary models do not provide. They can be considerably more cost-efficient and can additionally offer transparency and flexibility that closed source models cannot match.

Meta is releasing the new model now via a variety of different API platforms.