Meta CEO Mark Zuckerberg has intensified his call for a more open artificial intelligence ecosystem, releasing a new open-weight AI model, Muse Glimmer, designed to run directly on consumer devices. The 30-billion-parameter dense model, which fits on a laptop or single GPU, is being billed as one of the highest-performing models of its size, with weights freely available for download, customization, and local deployment.
The launch was accompanied by a 6,500-word manifesto titled “The Future is for Everyone: The Path to a Positive AI Future,” in which Zuckerberg argued that the greatest risk from advanced AI is not rogue machines but the centralization of superintelligence under a single government or corporation. He warned that concentrating such power “will naturally lead to outcomes that are less favorable for everyone else” and stressed that “there is no such thing as a singular benevolent superintelligence.”
Meta confirmed that Muse Glimmer was built through distillation from its larger Muse Spark system, with a version of Muse Spark 1.2 expected to be released in the coming weeks. The company had briefly paused open-weight releases for a strategy review but has now resumed them under the revived Meta Superintelligence Labs.
Zuckerberg’s essay also outlined a vision based on individual empowerment, invention over automation, and a balance of power for safety. To address safety concerns, Meta announced an independent board with authority to approve safety criteria for model releases, as well as a $1 billion “Future is for Everyone Fund” targeting U.S. communities hosting Meta data centers.
The debate around AI openness intensified as the lines between closed models, open-weight models, and truly open-source AI were drawn. Closed models like ChatGPT and Gemini keep weights private, while open-weight models like Muse Glimmer offer the weights but withhold training code, datasets, and methodology. True open-source AI would require full reproducibility. Stanford’s James Landay noted that open weights answer “Can I run this?” but open source answers “Can I trust this, improve it, and build the next thing on top of it?”
