Mark Zuckerberg, CEO of Meta, has called for the United States to lower barriers for open-source AI models to better compete with Chinese companies. This announcement came as Meta unveiled its new AI model, Muse Glimmer, on Monday (August 10). Muse Glimmer is designed to run directly on devices like a Mac or PC, using a single graphics card, making it smaller and more accessible than other leading AI models.

Zuckerberg emphasized the need for new policies to support open-weight models, which have publicly accessible components for customization. He argued that current U.S. restrictions give an advantage to Chinese competitors, who are advancing rapidly in open development. "Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data," Zuckerberg said.

Open-weight models are generally cheaper and offer more flexibility than closed models, which companies keep under tight control. As businesses grow wary of high AI costs and recent cybersecurity incidents, open-weight AI is gaining traction. Meta's launch of Muse Glimmer is part of its strategy to regain a strong position in the AI race, following the formation of a new superintelligence team in 2025.

Zuckerberg also advocated for AI model distillation, using powerful systems to train smaller models, and announced a governance structure to oversee safety criteria for future models. Meta plans to release more open-weight versions of its flagship models soon.

The U.S. government's stance on open-weight AI models remains cautious, as President Donald Trump's administration recently decided not to subject these models to voluntary safety tests. Meanwhile, Chinese startups like Moonshot and Alibaba continue to lead in open-weight AI development, posing a challenge to U.S. companies like OpenAI and Anthropic, whose models remain closed source.