Apple and Meta Team Up to Run 816-Billion-Parameter AI Model Across Dual Mac Studios
Meta's open-weight Muse Spark Open pairs with Apple's MLX framework to execute large-scale agentic workflows locally across two M5 Ultra desktops.
Apple and Meta have collaborated to enable Meta's new open-weight artificial intelligence model, Muse Spark Open, to run locally across paired Apple desktops, as reported by Mashable Tech (https://mashable.com/tech/apple-meta-muse-spark-open-mac-studio).
Announced on Oct. 15, the release permits developers to download and customize the 816-billion-parameter system rather than accessing it through hosted cloud APIs. However, the desktop setup requires substantial computing power: Meta's featured configuration links two M5 Ultra Mac Studios, each configured with up to 512GB of unified memory, running on Apple's open-source MLX machine-learning framework. Meta noted in a blog post that the linked configuration allows two Mac Studios to operate like a single supercomputer.
The technical alignment signals a thaw between the two companies following years of platform friction, including Meta chief executive Mark Zuckerberg's January 2025 appearance on Joe Rogan's podcast criticizing Apple's platform rules and pace of innovation. Speaking to Axios, Meta AI chief Alexandr Wang welcomed the joint effort, remarking that the team was pleased to see both companies cooperating.
Muse Spark Open expands on Meta's Muse Spark model family first introduced in April. Meta claims the architecture is the most capable U.S. open-weight model available, based on its internal vendor evaluations. Designed for multi-step reasoning, coding, and tool-assisted agentic tasks, Meta reported testing the architecture via a tool called Muse Code to modify the open-source Mastodon client Ice Cubes and automate video tasks in DaVinci Resolve, such as reviewing footage and reframing clips vertically.
The setup also addresses user privacy by design. Because processing takes place entirely on local hardware, Wang told Axios that Meta has no visibility into how developers query or run the model.
The release follows Muse Glimmer, a smaller variant built for individual laptops and PCs. Muse Spark Open and the requisite MLX Adapter are currently available for download on GitHub.
Sources
Written by
The Company Wire
Inside the companies building what’s next. Reporting on startups, technology, funding and the people shaping them.
