Meta Releases Muse Spark Open, an 816-Billion-Parameter Open-Weight Model

Meta on Thursday released Muse Spark Open, an open-weight AI model with 816 billion parameters that the company describes as the most capable open-weight model in the United States, according to TechCrunch.

The model has a context window of 1 million tokens and is natively multimodal, meaning it handles more than text. Meta says it is built for complex coding and agentic work, and that its own evaluation results show it outperforming other popular open-weight models. Those benchmark claims come from Meta and have not been independently confirmed.

What it runs on

Muse Spark Open is designed to run locally on Apple computers. Engadget reported that it needs at least 1 TB of memory, and Meta says it performs best on a Mac Studio with Apple's M5 Ultra chip.

• Parameters: 816 billion
• Context window: 1 million tokens
• Modality: natively multimodal
• Local target: Mac Studio with M5 Ultra, 1 TB or more of memory
• Availability: several API platforms, which TechCrunch did not name

Demonstrations

Meta showed the model paired with Muse Code, its AI coding terminal. In one demo it made bug fixes and added features to a Mastodon client, then ran a full build. In another it reviewed and cropped footage in the video editor DaVinci Resolve.

Why it matters

Open-weight models let customers run and inspect a system themselves, which can lower costs and give more flexibility than closed models. Running a model of this size on a desktop machine rather than a data center rack would be a notable shift for local AI work.

TechCrunch's report did not cover license terms, pricing or specific benchmark scores. Other coverage has noted that the license is not fully open, so developers should read the terms before building on the model.