Exhibitor login
AI Insider 16 October 2026

Meta launches Muse Spark Open: free AI model for Mac

Meta launches Muse Spark Open: free AI model for Mac

On October 15, 2026, Meta released the AI model Muse Spark Open, which is available for free download. This model is unique as it operates on just two Mac Studios, making it more accessible for developers and businesses. According to Meta, Muse Spark Open is the most powerful open model developed by a U.S. company to date, with 816 billion parameters capable of processing texts up to 1 million tokens.

To run Muse Spark Open, one needs two Mac Studios with M5 Ultra chips and 512 GB of memory each, resulting in a total of one terabyte of memory. This enables the use of a model of this scale in a desktop environment, which is advantageous for programming and long tasks performed by AI agents. The demos show that the model can even enhance complex applications and edit videos in DaVinci Resolve without an internet connection.

While Muse Spark Open delivers promising performance, Meta's claim that American open models are superior can be challenged. In performance comparisons with Chinese open models, such as Kimi K3 and GLM 5.3, Muse Spark Open ranks lower on some programming tasks. However, in the test for AI agents using external tools, the model tops the chart with 89.2 points. Users of the model must adhere to specific guidelines, such as mentioning ‘Built with Muse Spark Open’ for products utilizing the model.

For European companies, it is positive that the license permits usage in Europe, with agreements needing to be made with Meta Platforms Ireland. However, the required hardware is costly; a Mac Studio with M5 Ultra starts at €6,649, and two devices can quickly exceed €20,000. While the development of proprietary models on personal hardware becomes more attractive for companies concerned about data storage in the U.S. cloud, the high price of the required hardware currently poses a barrier. This demonstrates that Meta takes the open-source route seriously, although access to expensive hardware may hinder adoption.

Read the full article from AI Insider.