According to Reuters, Mark Zuckerberg, CEO of Meta Platforms, recently called for the United States to reduce restrictions on open-source artificial intelligence in order to better compete with Chinese competitors. Meanwhile, the social media giant released a new open-weight model on Monday and stated that more models will be introduced soon.
This new model, named Muse Glimmer, is smaller in size compared to leading artificial intelligence models of its competitors. It is designed to perform intelligent tasks on Macs or personal computers equipped with only a single graphics card. Its purpose is to meet the need for artificial intelligence systems that can be run directly on personal devices.
"We will soon launch a much larger model." Zuckerberg said in a video released alongside a 14-page article titled "The Future is for Everyone." In the article, he advocates for promoting AI's widespread use, rather than allowing it to be controlled by a select few.
"The belief that artificial intelligence is dangerous and only concentrating power at a high level is the only safe path, this view itself seems to have problems."
Zuckerberg's statement is the latest public endorsement of open-source artificial intelligence. As companies become more wary of the growing costs associated with artificial intelligence, and concern about recent cybersecurity incidents involving Anthropic, OpenAI, and Meta models, open-source artificial intelligence is becoming increasingly popular.
OpenWeighted models generally have lower costs compared to leading models from companies like OpenAI and Anthropic. They also provide their core components openly, allowing users to customize them—this is different from closed-source models that are completely controlled by companies.
After the uncontrolled OpenAI model invaded the artificial intelligence code collaboration platform Hugging Face, such restrictions began to draw attention. Due to the limitations of closed-source models in terms of network security, Hugging Face was forced to use a Chinese open-source model to counter this attack.
Chinese startup companies are currently leading the competition in open-weight models. The Kimi K3 from Moon Darkness, Alibaba's Qwen3.8-Max, and DeepSeek's V4-Flash all have performance capabilities that can compete with top American systems. In contrast, the leading models developed by American developers OpenAI, Anthropic, and Google under Alphabet are all closed-source weight-based models.
Meta also plans to disclose the model weights of Muse Spark 1.2. This is the company’s most advanced model, developed by a super-intelligent team that was formed last year with significant investment. Meta hopes to regain its dominant position in the artificial intelligence competition through this model.
The company has long supported open models, and was previously one of the few tech companies in the US that released such models. However, last year, the poor response to Llama 4 forced Meta to change its strategy.
In this article that covers a wide range of topics related to artificial intelligence, Zuckerberg also announced the establishment of a new $1 billion fund. By providing support to communities, he aims to alleviate concerns regarding the impact of Meta’s large-scale data center construction.
The growing opposition from local communities in the United States to data centers has become an important election issue. It is also one of the biggest obstacles faced by large tech companies in building the infrastructure needed for artificial intelligence.
Compared with countries such as China, the United States faces a significant disadvantage in that it is more difficult to construct basic infrastructure here. This was stated by Mark Zuckerberg. Meta plans to invest up to $14.5 billion in artificial intelligence infrastructure this year.
Zuckerberg also said that if the United States wants its companies to lead in the field of open weighted models, it is necessary to re-examine relevant policies, including those related to data use and "distillation." Distillation refers to the use of the output of a powerful artificial intelligence system to train smaller models, allowing the latter to perform some of the same tasks with much less computational power.
He stated that Meta will establish a governance structure that gives company independent directors the authority to approve safety standards for model releases.
When talking about open-source models, Zuckerberg said, "Currently, foreign laboratories have several advantages in this area, because American laboratories must comply with many additional restrictions regarding training data."
Zuckerberg said, “If we want the United States to remain ahead in open-source models in the long term, American policies must reduce these additional obstacles.” He also added that restricting people from using foreign open-source models is not a viable solution.
According to two sources familiar with the discussions, the Trump administration informed artificial intelligence developers earlier this month that the government would not require open-source artificial intelligence models to undergo voluntary safety tests.
