Meta Unveils Muse Spark Open: A High‑Performance Open‑Weight AI Built for Macs

Technical Specifications and Performance Claims

Meta describes Muse Spark Open as the most capable open‑weight model it has released in the United States. The system, named Muse Spark Open, is distinguished by its design for Mac operation and features a 816 billion‑parameter architecture with a one‑million‑token context window, which Meta says positions it to outperform other popular open‑weight models. Evaluation results released by Meta on Thursday back up these performance claims.

Mac‑Centric Design and Hardware Requirements

Muse Spark Open is natively multimodal and engineered for complex coding tasks and agentic labor. It has also been designed to run locally on Apple’s computers. Running it locally, however, demands substantial resources; Engadget reports a minimum requirement of one terabyte of RAM. Meta recommends the Mac Studio equipped with the M5 Ultra chip for optimal operation.

Still, if you can muster the memory, Muse Spark Open may prove a significant workhorse for coding and other desktop tasks.

Practical Deployments and Real‑World Applications

To showcase the model’s capabilities, Meta paired it with Muse Code, its AI coding terminal, and deployed it across several agentic workflows. One deployment involved fixing bugs, adding new features, and completing a full build for a Mastodon client.

bug fixes, new features, [and] full build

The company also deployed the model within DaVinci Resolve, the high‑end video editing software, where it demonstrated the ability to review and crop footage.

Benefits of Open‑Weight Models

Open‑weight models offer a range of distinct industry advantages not available to closed and proprietary alternatives, and they can be markedly more cost‑efficient while also delivering transparency and flexibility that closed‑source models do not provide.

Availability

Meta has made Muse Spark Open accessible immediately through multiple API platforms, enabling developers and enterprises to integrate the model into their applications without delay.