Menlo Park: Meta has launched a new open-weight artificial intelligence model as CEO Mark Zuckerberg called for the United States to reduce barriers facing domestic developers, arguing that American companies need stronger support to compete with Chinese rivals in the growing open-weight AI market.
The new model, Muse Glimmer, is significantly smaller than leading models developed by rival AI companies. Meta said it is designed for agentic tasks and can run directly on a Mac or PC using a single graphics card.
The approach reflects growing interest in AI models that can operate directly on users’ devices rather than relying entirely on expensive cloud-based systems. Zuckerberg said the launch comes at an important moment for the AI industry, with businesses increasingly concerned about the high costs of operating advanced AI models and growing cybersecurity risks linked to autonomous AI systems.
Meta has invested heavily in artificial intelligence as it seeks to strengthen its position in the increasingly competitive AI race. The company established a new superintelligence team last year as part of its efforts to accelerate AI development.
Zuckerberg used the launch to argue that US policies should be reconsidered to help American companies compete in open-weight AI. Chinese AI developers have gained significant momentum in the field, with models from companies including Moonshot, Alibaba and DeepSeek increasingly competing with systems developed by leading US AI companies.
Models such as Moonshot’s Kimi K3, Alibaba’s Qwen3.8-Max and DeepSeek’s V4-Flash have attracted attention for offering strong performance while making their core components available for wider use and customisation. By comparison, leading models from OpenAI, Anthropic and Google remain largely closed.
Zuckerberg argued that US companies face additional restrictions, particularly around the data they can use to train open-weight models. He said reducing those additional barriers would be important if the US wants its AI developers to remain competitive in the long term.
Open-weight models have gained popularity as companies look for alternatives to expensive frontier AI systems. Unlike fully closed AI models, open-weight systems provide access to key components that allow developers and businesses to customise models and adapt them for specific applications.
They can also be deployed on private infrastructure or compatible consumer devices, potentially reducing reliance on cloud-based AI services. That flexibility has become particularly important as companies look for ways to control AI costs and address concerns about data privacy and cybersecurity.
Recent incidents involving autonomous AI systems have also intensified discussions around how advanced models should be controlled. Anthropic, OpenAI and Meta have all reported incidents in which AI models accessed external systems during cybersecurity testing after safeguards or testing environments failed to contain them properly.
Hugging Face, an AI development and collaboration platform that was targeted by a rogue OpenAI model, previously said it had used a Chinese open-weight model to help defend against the attack. The company said closed models can impose restrictions on cybersecurity applications.
The Trump administration has also been reviewing how advanced AI models should be regulated. Earlier this month, US officials told AI developers that open-weight models would not be included in voluntary government safety testing, according to people familiar with the discussions.
Meta plans to introduce a governance structure under which independent directors will have authority to approve safety criteria before certain AI models are released, Zuckerberg said.
With the launch of Muse Glimmer, Meta is positioning itself not only as a competitor in the race for powerful AI systems but also as a major supporter of models that can be customised and run more efficiently outside large cloud platforms.
TRENDING | Taylor Swift’s music removed from Trump-linked social media posts
