Meta releases Muse Spark Open, an 816B model designed for Macs
Meta has released Muse Spark Open, an open-weight AI model with 816 billion parameters and a one-million-token context window. The company describes it as its most capable open-weight model in the United States and says it was designed to run locally on Apple Macs.
Muse Spark Open is natively multimodal and intended for complex coding tasks and agentic work. Meta is making the model available through several API platforms, while also presenting local Mac deployment as a key part of its positioning.
Large context and demanding local hardware
Meta published evaluation results claiming that Muse Spark Open outperforms other popular open-weight models. The release combines a very large parameter count with a one-million-token context window, specifications aimed at workloads that require substantial amounts of code or other material to remain in context.
Running the model on a Mac locally nevertheless requires substantial capacity. Engadget reported a minimum requirement of 1TB of RAM, and the model is expected to perform best on a Mac Studio equipped with Apple’s M5 Ultra chip. That requirement limits the practical local audience despite the model’s Mac-focused design.
Testing Muse Spark Open in agentic workflows
Meta paired the model with Muse Code, its AI coding terminal, and deployed it in agentic workflows. In one example, it used the system on a Mastodon client for bug fixes, new features and a full build. The release extends Meta’s work in AI-assisted development, including Meta’s agentic coding model release and its focus on agentic coding tasks.
Meta also said it deployed Muse Spark Open to run DaVinci Resolve, the high-end video-editing application. In that test, the model reviewed and cropped footage, illustrating the company’s multimodal use-case claims beyond software development.
What open weights change for buyers
Open-weight models can offer cost efficiency, transparency and flexibility that closed, proprietary systems do not provide. Those attributes may matter to organizations that need more control over model deployment or want to adapt a model to specific internal workflows.
For businesses, Muse Spark Open’s practical implication is that local AI capability must be weighed against infrastructure requirements: teams should compare the cost of memory and suitable hardware with the coding, automation and media workflows they expect the model to support.