According to a report by Reuters, Meta Platforms CEO Mark Zuckerberg recently urged the U.S. to lower restrictions on open-source artificial intelligence to better compete with Chinese rivals. Meanwhile, the social media giant released a new open-weight model on Monday and hinted that more models are on the way soon.
The new model, named Muse Glimmer, is smaller than the leading AI models from competitors. It’s mainly designed to handle agent tasks on a Mac or personal computer with just a single graphics card, aiming to meet the growing demand for AI systems that can run directly on personal devices.
“We’ll be releasing larger models soon,” Zuckerberg said in a video accompanying his 14-page essay titled “The Future is for Everyone.” In the piece, he advocates for democratizing AI rather than concentrating it in the hands of a few.
“The notion that AI is so dangerous that the only safe path is to centralize power seems fundamentally flawed,” he argued.
Zuckerberg’s remarks mark the latest public endorsement of open-weight AI. This approach is gaining traction as companies grow increasingly wary of ballooning AI expenditures and recent cybersecurity incidents involving models from Anthropic, OpenAI, and Meta.
Open-weight models are typically more cost-effective than the leading models from companies like OpenAI and Anthropic. They also make their core components publicly available for customization, unlike closed-source models that are fully controlled by corporations.
These restrictions came under scrutiny after the AI code collaboration platform Hugging Face was breached by an out-of-control OpenAI model. Because closed-source models have limitations when it comes to cybersecurity use, Hugging Face had to rely on a Chinese open-weight model to fend off the attack.
Chinese startups are currently leading the race in open-weight models. Moonshot AI’s Kimi K3, Alibaba’s Qwen3.8-Max, and DeepSeek’s V4-Flash all perform at levels comparable to top American systems. In contrast, leading U.S. developers like OpenAI, Anthropic, and Alphabet’s Google all use closed-weight models.
Meta also plans to release the model weights for Muse Spark 1.2. This is the company’s most advanced model to date, developed by the superintelligence team it assembled with heavy investment last year. Meta hopes this will help it regain a competitive edge in the AI race.
The company has long championed open models and was previously one of the few U.S. tech firms to release such models. However, the lukewarm reception to Llama 4 last year forced Meta to rethink its strategy.
In the wide-ranging essay on AI topics, Zuckerberg also announced a new $1 billion fund to ease concerns about the impact of Meta’s massive data center buildout by providing support to local communities.
Growing local opposition to data centers in U.S. communities has become a significant election issue and one of the biggest hurdles for tech giants seeking to build the infrastructure needed to advance AI.
“One clear disadvantage the U.S. faces compared to countries like China is that building infrastructure here is much harder,” Zuckerberg said. Meta plans to invest up to $145 billion in AI infrastructure this year.
Zuckerberg also said that if the U.S. wants its companies to lead in open-weight models, it needs to revisit relevant policies, including those on data usage and “distillation.” Distillation involves using the output of a powerful AI system to train smaller models, allowing them to perform some of the same tasks with far less computing power.
He added that Meta will establish a governance structure that empowers the company’s independent directors to approve safety standards for model releases.
Speaking about open-source models, Zuckerberg noted: “Foreign labs currently hold certain advantages here because U.S. labs have to comply with more additional restrictions on training data.”
“If we want American open-source models to stay ahead in the long run, U.S. policy needs to reduce these extra hurdles,” Zuckerberg said. He also added that restricting people’s access to foreign open-source models isn’t an effective solution.
According to two sources familiar with the discussions, the Trump administration told AI developers earlier this month that the government will not require voluntary safety testing for open-weight AI models.
