Meta launches Muse Spark Open, an 816-billion-parameter open-weight model that can run on Macs

Meta has unveiled Muse Spark Open, an open-weight AI model that the company describes as the most capable in its class in the United States. According to TechCrunch, which reported the launch on October 15, 2026, the model is designed to run locally on Apple computers.

According to the article, the model has 816 billion parameters and a context window of 1 million tokens. Citing evaluation results it published on Thursday, Meta claims it outperforms other popular open-weight models. These figures come from the company itself, and the article does not specify which benchmarks were used or which models it was compared against.

Muse Spark Open is natively multimodal and is built for complex coding tasks and agentic work. Its ability to run locally comes with a major trade-off: according to Engadget, cited by TechCrunch, it requires at least 1 terabyte of RAM. The model would run best on a Mac Studio equipped with the M5 Ultra chip. So although it runs on Macs, in practice it is limited to very high-end setups.

Meta says it paired the model with Muse Code, its AI coding terminal, and tested it on several agentic workflows. These included code fixes for a Mastodon client, which according to the company involved bug fixes, new features and a full build. It also applied the model to operating DaVinci Resolve, the professional video editing software, where the model was able to review and trim footage. These examples come from Meta itself and have not been independently verified.

The article notes that open-weight models offer advantages over closed, proprietary ones. They can be considerably cheaper and offer a transparency and flexibility that closed models do not allow. Meta is already offering the new model through several API platforms.

The downloaded text is short and does not include details on licensing, pricing or specific evaluation results, so this summary is limited to what the article reports.