Meta has released Muse Glimmer, the first open model from its new Superintelligence Labs. This 30B agent model runs on consumer hardware once weights are compressed, requiring less than 20 GB of memory. Mark Zuckerberg wrote an accompanying essay defending the distillation of other labs’ models and calling for fewer restrictions on US companies. This stance acts as a direct counterpunch against OpenAI and Anthropic. An open-weight version of Muse Spark 1.2 should follow soon, according to the Wall Street Journal.
The move signals a strategic shift away from closed ecosystems towards a model where compute power becomes a tradable commodity. By releasing weights that run locally, Meta reduces reliance on cloud infrastructure and challenges the dominance of large tech firms. The plan to sell compute by auction suggests a future where processing power is allocated through market mechanisms rather than subscription fees.
* Muse Glimmer operates on consumer hardware with under 20 GB of memory
* Open-weight Muse Spark 1.2 is expected to launch in the near future
* Zuckerberg advocates for reduced regulatory restrictions on US AI laboratories
