Meta releases Muse Spark Open, a model built to run on Macs

Meta has unveiled what it calls the "most capable" open-weight model in the United States. Muse Spark Open also stands out for another reason: it was designed to run on Macs.

The model draws on 816 billion parameters and a one-million-token context window. According to evaluation results Meta published on Thursday, it outperforms other popular open-weight models.

Multimodal and built for agents

Muse Spark Open is natively multimodal and built for complex coding tasks and agentic work. It can run locally on Apple's computers — though you'll need serious hardware: at least one terabyte of RAM, per Engadget. The model performs best on a Mac Studio equipped with the M5 Ultra chip.

If you can assemble the memory, the model may prove a capable workhorse for coding and other desktop chores. Meta paired it with Muse Code, its AI coding terminal, and put it through various agentic workflows, including work on a Mastodon client that involved bug fixes, new features, and a full build.

The company also set it loose on DaVinci Resolve, the high-end video editing software, reporting that the model could review and crop footage.

Why open weights matter

Open-weight models offer the industry advantages that closed, proprietary systems can't match. They tend to be far more cost-efficient, and they provide a degree of transparency and flexibility that closed models withhold.

Meta is making the new model available now through a number of API platforms.

Source: TechCrunch