Meta Platforms CEO Mark Zuckerberg has advocated for reduced restrictions on open-source AI in the U.S. to enhance competition with Chinese counterparts. The company introduced a new, more compact and customizable AI model named Muse Glimmer, tailored for running tasks on personal computers with a single graphics card. Zuckerberg mentioned in a video post accompanying a written piece that larger models are in the pipeline.
Zuckerberg emphasized the importance of democratizing AI rather than centralizing power in the hands of a few. The move towards open-weight AI has gained momentum as companies seek to cut costs and address security concerns following incidents involving models from Anthropic, OpenAI, and Meta.
The spotlight has turned to open-weight models, particularly after the emergence of notable Chinese models like Kimi K3 and instances where these models outperformed closed-source alternatives in cybersecurity scenarios. Meta, along with other tech giants, endorsed open-weight AI in a recent joint statement.
Chinese startups have taken the lead in open-weight AI development, with models like Kimi K3, Qwen3.8-Max, and V4-Flash rivaling top U.S. systems. In contrast, leading U.S. models from developers like OpenAI, Anthropic, and Google are closed weight.
Meta’s stock, which has experienced a 10% decline this year, saw a 3% increase in premarket trading following the announcement. The company plans to release Muse Spark 1.2, its most advanced model, to regain its competitive edge in the AI sector.
Zuckerberg also unveiled a $1 billion fund aimed at addressing concerns surrounding the expansion of Meta’s data centers and their impact on communities. He stressed the need for the U.S. to rethink policies to foster the development of open-weight models and ensure data use transparency and efficiency.
To support the advancement of open-source AI models, Meta will implement a governance structure empowering independent directors to approve safety criteria for model releases. The U.S. government, under President Trump’s administration, has steered clear of imposing safety tests on open-weight AI models, opting for a more hands-off approach.