Meta has introduced Muse Spark Open, its latest open-weight AI model, which is notable for being designed to run locally on Apple silicon-powered Macs. The model, launched on October 15, 2026, incorporates an impressive 816 billion parameters and supports a 1 million token context window, positioning it as one of the most powerful open-weight models available in the U.S. Meta highlights that this model outperforms other open-weight models based on their evaluation metrics.

Muse Spark Open is a multimodal AI designed to handle complex tasks, including sophisticated coding and agentic workflows. Meta has integrated the model with Muse Code, its AI coding terminal, and demonstrated its capabilities by automating bug fixes and feature additions for a Mastodon client. The AI also assisted with tasks like reviewing and cropping videos in DaVinci Resolve, showcasing its practical application in high-end video editing.

Running the model locally demands significant hardware resources, with a minimum requirement of 1 terabyte of RAM. Meta recommends the Mac Studio equipped with the M5 Ultra chip as the optimal machine to support this demanding workload. While the technical requirements are high, those who can leverage this model will gain a powerful tool for desktop-based coding and other professional tasks.

Meta is releasing Muse Spark Open through various API platforms, making it accessible to developers and enterprises interested in the advantages of open-weight models. These models offer greater transparency, flexibility, and cost-efficiency compared to proprietary AI systems, giving users more control over customization and deployment. This launch reflects Meta’s commitment to expanding AI’s capabilities on local devices in collaboration with Apple.