Meta launches an open-weight AI model that runs on Macs

Your coding assistant on a Mac could soon be significantly more powerful and less expensive to run.

Meta has introduced Muse Spark Open, an open-weight AI model designed to run locally on Mac computers. This model, featuring 816 billion parameters and a 1 million token context window, claims superior performance over existing open-weight alternatives, according to Meta's internal evaluations. Muse Spark Open is natively multimodal and engineered for complex coding and agentic tasks. While it requires substantial hardware, specifically at least 1 terabyte of RAM, it is optimized for Apple's Mac Studio with the M5 Ultra chip. Meta is making the model accessible via various API platforms. The company highlights that open-weight models offer cost efficiency, transparency, and flexibility compared to proprietary AI solutions.

Meta's release of Muse Spark Open signifies a significant push towards democratizing powerful AI capabilities by enabling local execution on consumer hardware like Macs. The emphasis on an open-weight model suggests a strategic move to foster wider adoption and community development, potentially accelerating innovation in AI applications that require on-device processing. This could lead to new categories of AI-powered tools and services that are more private and customizable for end-users.

The requirement for high-end hardware, such as 1TB of RAM and M5 Ultra chips, indicates that while the model is accessible, achieving peak performance will still be a premium experience. This could create a bifurcated market where advanced AI tasks are relegated to users with top-tier machines, while broader adoption on less powerful Macs might be limited. Developers will need to balance model efficiency with hardware constraints when building applications that leverage Muse Spark Open.