Model Overview
Meta has launched what it describes as its most capable open-weight model in the United States. Named Muse Spark Open, the system is engineered to integrate with Mac desktop environments.
The model features 816 billion parameters and a 1 million token context window. According to evaluation benchmarks published by Meta, it delivers performance metrics surpassing several existing open-weight alternatives.
Hardware Requirements
Muse Spark Open is built natively for multimodal processing, complex software engineering, and agentic workflows. It supports local execution on Apple computers, though hardware demands remain steep: running the model locally requires at least 1 terabyte of RAM. Optimal performance is targeted for the Mac Studio equipped with the M5 Ultra chip.
Capabilities and Use Cases
When paired with Muse Code, Meta's artificial intelligence coding terminal, the model can execute agentic workflows such as implementing bug fixes and building new features for software applications.
Meta also demonstrated the model running within DaVinci Resolve, a professional video editing suite, where it assisted with reviewing and cropping footage.
Industry Context
Open-weight artificial intelligence models offer several advantages over proprietary systems, including cost efficiency, greater architectural transparency, and deployment flexibility.
The new model is currently available through multiple application programming interface platforms.
