Meta Launches AI Model, Calls for Fewer Rules

Meta Releases Open-Weight AI Model, Zuckerberg Seeks Reduced Data Restrictions

Meta CEO Mark Zuckerberg called for changes to U.S. policies on training data and import limits for open-weight AI models to improve competition with Chinese developers. The company released a new open-weight model on Monday and stated plans to introduce additional models in the coming months.

The model, Muse Glimmer, contains fewer parameters than leading systems from other companies and targets agentic tasks. It is designed to operate on a Mac or PC equipped with one graphics card, addressing interest in AI systems that function on local devices without cloud connections.

Open-weight models generally cost less to operate than those from companies such as OpenAI and Anthropic. Their core components are publicly available, allowing users to modify them, in contrast to closed models whose internal details remain proprietary.

The release occurs as Meta works to expand its AI capabilities following the creation of a superintelligence team in the prior year. Zuckerberg’s comments represent continued backing for open-weight approaches amid business concerns over rising operational expenses and security events tied to models from multiple providers.

Hugging Face reported last month that it employed a Chinese open-weight model to counter an attack from a compromised OpenAI system, citing usage limits on closed models for security applications. Meta shares, which declined roughly 10 percent year to date, rose 1 percent in premarket trading.

Zuckerberg stated that current U.S. rules on training data place domestic developers at a disadvantage relative to foreign labs. He argued that policy adjustments are required for American open-weight models to maintain leadership and that blocking access to foreign open-weight systems would not resolve the issue.

The Trump administration informed AI companies earlier this month that open-weight models would not face voluntary safety evaluations, according to two people familiar with the matter. Zuckerberg also endorsed model distillation, in which a larger system generates training data for a smaller one, and said Meta would establish oversight allowing independent directors to review safety standards before releases.

Chinese developers have introduced several open-weight models with performance comparable to leading U.S. closed systems. Moonshot’s Kimi K3, Alibaba’s Qwen3.8-Max, and DeepSeek’s V4-Flash are among those cited. In comparison, current top models from OpenAI, Anthropic, and Alphabet’s Google remain closed source. Meta indicated it will continue releasing open-weight models while implementing the described governance process.