Meta's Muse Spark Open Brings Frontier AI to Local Hardware
AI-generated image · Bay Street Wire
The launch of a massive open-weight model designed for Mac Studio signals a shift toward local execution for complex creative and coding workflows.
Meta has introduced Muse Spark Open, which the company describes as the most capable open-weight model in the U.S. As TechCrunch first reported, the model features 816 billion parameters and a 1 million token context window, designed specifically to operate locally on Apple computers.
While the model is available via various API platforms, its primary distinction is the ability to run on consumer hardware. TechCrunch notes that Meta paired the model with Muse Code, its AI coding terminal, to execute agentic workflows such as implementing new features and bug fixes for a Mastodon client. Additionally, Meta demonstrated the model's utility in high-end media production by using it to review and crop footage within the video editing software DaVinci Resolve.
Local execution requires significant hardware resources. As reported by TechCrunch via Engadget, the model requires a minimum of 1 terabyte of RAM and performs optimally on the Mac Studio equipped with the M5 Ultra chip.
By releasing an open-weight model rather than a proprietary one, Meta is offering a path toward greater transparency and flexibility. TechCrunch reports that these models can also be significantly more cost-efficient than closed-source alternatives, potentially reducing the reliance on centralized platform costs for complex desktop tasks and agentic labor.