Meta Platforms CEO Mark Zuckerberg has urged for reduced barriers on open-source AI in the U.S. to enhance competition with Chinese counterparts. The company unveiled a smaller and more customizable AI model named Muse Glimmer, tailored for running tasks on individual computers with a single graphics card. Zuckerberg expressed plans for larger models in the pipeline to be released soon.
Advocating for wider dissemination of AI and cautioning against concentrated power, Zuckerberg emphasized the benefits of open-weight AI models over closed systems, citing recent cybersecurity incidents involving models from Anthropic, OpenAI, and Meta. Open-weight models, which allow customization by users and are cost-effective, are gaining prominence in contrast to closed models maintained by companies.
The rise of Chinese open-weight models, such as Kimi K3 from Moonshot and Qwen3.8-Max from Alibaba, has propelled Chinese startups ahead in the open-weight model arena. In response, Meta and other tech companies, including Nvidia and Microsoft, have voiced support for open-weight AI. Despite Meta’s stock decline this year, shares rose nearly 3% in premarket trading following the announcement.
Meta intends to release Muse Spark 1.2, its advanced AI model, developed by a dedicated superintelligence team. The company’s commitment to open models is evident from its previous releases and efforts to address community concerns. Zuckerberg unveiled a $1 billion fund to allay fears surrounding data-center expansion and emphasized the need for infrastructure development in the U.S.
Highlighting the necessity for policy changes to foster American leadership in open-weight models, Zuckerberg stressed the importance of data use and distillation. Meta plans to implement a governance structure empowering independent directors to approve model releases. Notably, the U.S. administration declined to subject open-weight AI models to voluntary safety tests, signaling ongoing developments in the AI landscape.