Meta CEO Mark Zuckerberg called on the U.S. to lower barriers for open-weight AI models so American developers can more effectively compete with Chinese ones. This statement came on August 10, when the social media giant unveiled a new open-weight model and announced plans to release more such models soon.
The new Muse Glimmer model is significantly smaller than leading AI models from competitors. It is designed for agent tasks and can run on a Mac or PC with a single graphics card. Thus, Meta aims to capitalize on the demand for AI systems that operate directly on users’ devices.
Open-weight models are generally cheaper than leading developments from so-called “advanced labs” such as OpenAI and Anthropic. They also provide public access to key components, making it easier to fine-tune models, unlike closed systems that companies fully control.
Meta’s launch comes as the company seeks to strengthen its position after creating an expensive new superintelligence development team in 2025. The goal was to bolster Meta’s standing in the competitive AI race. Zuckerberg’s statement also served as further evidence of support for open-weight models, which are gaining popularity amid rising company costs for AI and concerns over recent cybersecurity incidents involving models from Anthropic, OpenAI, and Meta.
The AI collaborative development platform Hugging Face, which was hacked by an unauthorized OpenAI model, stated in July 2026 that it used a Chinese open-weight model to defend against the attack. According to the company, closed models have limitations regarding their use for cybersecurity tasks.
Meta’s shares, which have fallen by approximately 10% this year, rose by 1% during pre-market trading on August 10.
U.S. policy on AI models
Zuckerberg noted in his statement that the U.S. needs to review its policy if American companies want to lead in developing open-weight models. Chinese startups already lead in this segment: Moonshot’s Kimi K3, Alibaba’s Qwen3.8-Max, and DeepSeek’s V4-Flash demonstrate results comparable to leading systems from American AI labs. Meanwhile, leading models from American developers OpenAI, Anthropic, and Google (Alphabet company) remain closed.
“Foreign labs currently have several advantages since American labs must comply with many additional restrictions on training data,” said Zuckerberg. “U.S. policy should reduce these additional barriers if we want American open-weight models to eventually lead,” he added, noting that restricting access to foreign open-weight models is not an effective solution.
The administration of U.S. President Donald Trump previously informed AI developers in August that open-weight models would not undergo voluntary safety tests, according to two sources familiar with the discussions.
In his statement, Zuckerberg also advocated for the distillation of AI models—using a powerful AI system to train a smaller model. He added that Meta would implement a governance system giving the company’s independent directors the authority to approve safety criteria before releasing models.
Source: The Straits Times
