Meta’s new Glimmer AI model offers a hint at Zuckerberg’s personal intelligence vision

Meta on Monday unveiled Muse Glimmer, an open-weight model that gives a much clearer look at how Mark Zuckerberg wants “personal superintelligence” to work in practice. The company is positioning it as the local, user-controlled version of its more capable closed model, Muse Spark, which arrived in April.

Glimmer has 30 billion parameters and is released under Apache 2.0, which means developers can download it and modify it. Meta says it can run on a Mac or PC with a single consumer GPU and handle multi-step agent work: calling tools, writing and debugging code, working with files and screenshots, and carrying a task through an extended workflow. It supports text and images and was trained across more than 100 languages.

The use cases Meta keeps pointing to are the intimate ones. Schedules. Messages. Files. The kind of work that only gets useful if an AI can see a lot of personal context. By keeping that processing on-device instead of shipping it off to the cloud, Meta is trying to make the whole thing feel more privacy-sensitive. Glimmer is also meant to be always on, and able to work with or without an internet connection.

That fits neatly with Zuckerberg’s own pitch. In a new letter on Monday, he said distributing superintelligence widely could open a new era of personal empowerment, with tools that work 24/7 on someone’s behalf and help with relationships, health, career, finances, home management, hobbies, and more. He also said the biggest promise is free or affordable access.

But Meta is drawing a line at the same time it’s talking about openness. Muse Spark stays closed-weight. Glimmer can be downloaded, fine-tuned, and run on a user’s hardware. So this isn’t just a product launch. It’s an early marker of where Meta is willing to let people actually own the AI, and where it plans to keep the more powerful stuff behind the company wall.