Overview
- Meta publicly released Muse Glimmer on Monday, an open-weight model whose trained parameters are available for download.
- Muse Glimmer is a compact, distilled model reported at about 30 billion parameters and engineered to run locally on a Mac or PC with a single GPU for agent-style tasks like scheduling and file management.
- Meta said it will also open the weights for a version of its larger Muse Spark (1.2), letting outside developers inspect and modify the model’s trained parameters.
- The company pledged governance changes that give independent directors the authority to approve safety criteria before model releases while Zuckerberg called for policy shifts favoring distillation and broader access.
- The move heightens a U.S.-China competition over open-weight models, ties into recent security incidents that have split opinion on openness, and could change costs and control for developers who want local, customizable AI.