Meta unveils new AI model as Zuckerberg pushes open-weight strategy against Chinese rivals

Dubai: Meta Platforms unveiled a new artificial intelligence model on Monday as Chief Executive Mark Zuckerberg urged the United States to reduce barriers to open-weight AI development, arguing that broader access to the technology is critical for competing with increasingly capable Chinese rivals. The new model, called Muse Glimmer, is designed to run agentic AI tasks on a Mac or PC using a single graphics card, targeting demand for AI systems that operate directly on users' devices rather than requiring large-scale cloud infrastructure. Muse Glimmer is smaller than some of the leading models developed by competitors, but Meta said additional, more powerful models are expected to follow. "We've got even bigger models that are coming soon," Zuckerberg said in a video accompanying a 14-page essay titled "The Future is for Everyone." In the essay, Zuckerberg argued against concentrating AI capabilities in the hands of a small number of companies. "The notion AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic," he said. Open-weight AI gains momentum Meta's latest move reflects growing support for open-weight AI models, which typically provide developers access to core model components and are generally cheaper to deploy and customise than closed systems developed by companies such as OpenAI and Anthropic. Support for open-weight AI has increased as businesses seek to reduce AI costs and address concerns over recent cybersecurity incidents involving AI models from several major developers. Chinese companies have emerged as leading competitors in the open-weight segment. Models including Moonshot's Kimi K3, Alibaba's Qwen3.8-Max and DeepSeek's V4-Flash have achieved performance levels that rival some of the most advanced US systems. By contrast, leading models from OpenAI, Anthropic and Google's parent company Alphabet remain closed-weight systems. New models planned Meta also said it plans to release the weights of Muse Spark 1.2, described as its most advanced model and developed by a superintelligence team established last year. The company had previously been one of the strongest advocates of open AI systems, although criticism of its Llama 4 models prompted a reassessment of strategy. Meta's shares, down roughly 10% this year, rose nearly 3% in premarket trading on Monday. Infrastructure and policy focus Alongside the AI announcements, Zuckerberg unveiled a $1 billion fund aimed at helping communities address concerns linked to the rapid expansion of AI-focused data centres. The issue has become increasingly contentious in the United States, where local opposition to data centre construction has emerged as a significant obstacle to technology companies' expansion plans. "One significant disadvantage that the U.S. has compared to countries like China is that it is more difficult to build infrastructure here," Zuckerberg said. Meta expects to spend up to $145 billion on AI infrastructure this year. Zuckerberg also called for a reassessment of US policies affecting AI development, particularly rules governing data access and distillation, a process that uses the output of large AI models to train smaller, more efficient systems. Governance changes Meta announced plans to implement a governance structure that will give independent directors authority to approve safety criteria for future model releases. According to Zuckerberg, US developers face competitive disadvantages because of restrictions not imposed on some foreign rivals. "Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data," he said. "U.S. policy must reduce this additional friction if we want American open source models to lead over time." He added that restricting access to foreign open-source models would not be an effective solution. The comments come after reports that the administration of President Donald Trump informed AI developers earlier this month that open-weight AI models would not be subject to voluntary safety testing requirements.