Meta has launched Muse Spark Open, an open-weight AI model boasting 816 billion parameters and a 1 million token context window. According to Meta, the model offers superior performance compared to other popular open-weight models. Developed in partnership with Apple, Muse Spark Open is natively multimodal and engineered for complex coding tasks and agentic workflows. A notable feature is its ability to run locally on Apple computers, though running it locally requires significant computing power, specifically at least 1 terabyte of RAM, and performs best on a Mac Studio with an M5 Ultra chip. Meta demonstrated the model's capabilities by pairing it with Muse Code, its AI coding terminal, to execute tasks such as bug fixes and new features for a Mastodon client. Additionally, it was deployed to run DaVinci Resolve, where it reviewed and cropped video footage. Open-weight models offer industry advantages over closed models, including cost-efficiency, transparency, and flexibility. Meta is currently making the new model available through various API platforms.
Meta launches an open-weight AI model that runs on Macs
Meta has released Muse Spark Open, an open-weight AI model featuring 816 billion parameters and a 1 million token context window. Designed in collaboration with Apple, the model is built to run locally on Apple silicon, specifically requiring at least 1 terabyte of RAM and performing best on a Mac Studio equipped with the M5 Ultra chip. Meta states the natively multimodal model is built for complex coding tasks and agentic labor.
Why it matters
The release highlights a collaboration between Meta and Apple to bring high-capacity, open-weight AI models directly to local desktop hardware. This development emphasizes the growing push for local hardware execution in agentic labor and complex coding workflows.