Meta releases open-weight model designed for local Mac execution
Meta has introduced an open-weight artificial intelligence model called Muse Spark Open, featuring an exceptionally large parameter count and context window.
2 min read·Friday, 16 October
Meta has introduced an open-weight model — meaning the internal weights and parameters are publicly shared for developers to inspect and run — called Muse Spark Open. Released on Thursday, the model is designed to handle complex coding tasks and autonomous agent workflows while offering native multimodal capabilities.
The model relies on 816 billion parameters — the adjustable variables that determine a model's complexity and capability — and supports a 1 million token context window, which is the amount of text the model can process at once. Evaluation results released by Meta suggest the system provides performance superior to other popular open-weight alternatives.
Unlike many large models that require cloud infrastructure, Muse Spark Open was specifically designed to run locally on Apple computers. However, practical local execution demands substantial hardware resources. Operating the model locally requires at least 1 terabyte of RAM, and the system is optimized to run on a Mac Studio equipped with the M5 Ultra chip.
A hardware setup featuring stacked Mac Studios and a connected display running Muse Spark Open locally. · Source ↗
Practical applications
Meta paired the model with Muse Code, its artificial intelligence coding terminal, to test its capabilities in real-world scenarios. The tests included handling bug fixes, building new features, and managing full builds for a Mastodon client.
Additionally, the company deployed the model to assist with DaVinci Resolve, a high-end video editing software. In these tests, the model successfully reviewed and cropped video footage.
Industry availability
Open-weight models provide distinct advantages over proprietary systems, offering greater cost-efficiency, transparency, and flexibility for developers. Alongside local hardware options, Meta is making the new model available through various application programming interface platforms.
Google AI Edge Gallery & Gemma 4 Setup GuideFree guide
Comments