Meta has launched what it calls its “highest performing” open-class model in the United States. Dubbed Muse Spark Open, this model also has the distinction of being designed to work with Mac.
The new model leverages 816 billion parameters and 1 million token context windows, according to evaluation results released by Meta on Thursday, which it claims provides better performance than other popular open-weight models.
Muse Spark Open is natively multimodal and built for complex coding tasks and agent work. It is also designed to run locally on Apple computers. However, running the model locally still requires a lot of power and requires at least 1 terabyte of RAM, according to Engadget. This model works best via Mac Studio with the M5 Ultra chip.
Still, if you can scrape up some memory, Muse Spark Open can provide significant horsepower for coding and other desktop tasks. Meta says it has combined this model with its AI coding terminal, Muse Code, and deployed it across a variety of agent workflows, including coding fixes for the Mastodon client, including “bug fixes, new features, and complete builds.”
The company said it also brought the model to its service running DaVinci Resolve, a high-end video editing software, and revealed that the model can review and trim footage.
Open weight models offer a number of unique industry advantages not available in closed and proprietary models. Open-weight models are much more cost-effective and can also provide transparency and flexibility not possible with closed-source models.
Meta is currently making new models available through various API platforms.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
