Meta is strengthening its support for openly available artificial intelligence with the release of Muse Glimmer, an open version of its powerful Muse Spark model. The move reflects CEO Mark Zuckerberg’s argument that advanced AI should be distributed widely rather than controlled by a small group of companies, tells The New York Times.
Muse Glimmer is nearly identical to Muse Spark and can generate text, code, and images. However, it is an open-weights model rather than fully open source. Its underlying model weights are publicly available, while its complete source code is not. Muse Spark, launched as a paid, closed model, will remain proprietary.
Alongside the release, Zuckerberg published a 14-page essay outlining his vision for AI. He argued that distributing increasingly capable AI could give individuals greater control over the technology while helping the United States maintain its technological lead over China. His position contrasts with approaches advocated by OpenAI and Anthropic, which have emphasized restrictions and controlled development because of potential AI risks.
The debate has intensified as Chinese companies release capable, inexpensive open AI models. These systems are gaining users in the United States and developing countries as businesses seek to reduce AI costs. Meta itself previously shifted toward closed models after falling behind competitors. Its Muse Spark model, developed under chief AI officer Alexandr Wang, currently trails leading models on several coding, reasoning, and writing benchmarks but is positioned as a lower-cost option.
Zuckerberg also envisions personal AI agents that continuously assist people with work, finances, relationships, and everyday tasks. He called for revised government oversight, defended model distillation, and announced a $1 billion fund for communities hosting Meta data centers.
Meta’s next major model, internally called Watermelon, is expected to outperform Muse Spark. Whether Meta will release it openly remains undecided.
