Mark Zuckerberg attacks ‘closed’ AI rivals as Meta returns to open models

San Francisco, Meta CEO Mark Zuckerberg called for the US to lower barriers for open-source AI models to compete with Chinese rivals as Meta unveiled a new open-weight model called Muse Glimmer on Monday. The compact model, designed for agentic tasks, can run on a Mac or PC with a single graphics card, targeting the demand for AI systems that operate directly on users' devices.

This move follows Meta's formation of a costly superintelligence team last year to regain its position in the competitive AI market. Zuckerberg's support for open-weight AI, which is gaining popularity due to cost concerns and cybersecurity worries over closed-source models from Anthropic, OpenAI, and Meta, reflects a growing trend among businesses wary of high AI expenses.

Hugging Face, an AI coding collaboration platform, recently used a Chinese open-weight model to counter a security threat from a rogue OpenAI model. Open-weight models, known for their affordability and customizable components, are currently outperforming top systems from US AI labs, such as OpenAI, Anthropic, and Alphabet's Google, which remain closed source.

Zuckerberg emphasized the need for US policy changes to enable domestic firms to excel in open-weight models, as Chinese startups are currently leading the race with models like Moonshot's Kimi K3, Alibaba's Qwen3.8-Max, and DeepSeek's V4-Flash. US policy should eliminate the extra friction faced by American open-source models, Zuckerberg stated, asserting that restricting foreign open-source models is not a viable solution.

The US administration ruled out subjecting open-weight AI models to voluntary safety tests earlier this month, according to sources familiar with the discussions. Zuckerberg also proposed AI model distillation, a process where a powerful AI system trains a smaller model, and promised to establish a governance structure for approving safety criteria for model releases.