From 2022, America has a solid lead in artificial intelligence with advanced models from leading companies such as OpenAI, Google DeepMind, Anthropic, and xAI. But a growing number of experts fear that the United States is starting to fall behind when it comes to creating open AI models that can be downloaded, customized and run locally.
Open models from Chinese companies such as Kimi, Zai, Alibaba and DeepSeek are now rapidly gaining popularity among researchers and engineers around the world, leaving the United States behind in the increasingly essential area of artificial intelligence innovation. “The United States needs open models to strengthen its advantage at every level of the AI stack” – Nathan Lambert, founder of the company Project ATOM (American really open models).says WIRED.
The most advanced models of US companies can only be accessed through a chatbot interface or by querying the companies’ servers via an application programming interface, or API. OpenAI and Google have released open-weight models, but they are much less powerful than Chinese offerings that are better suited to modification and offer greater support for developers. Chinese modelers also benefit from the open source of their models, as the best ideas and fixes from external researchers can be incorporated into future releases.
Lambert, who is also a researcher at the Allen Institute for AI (Ai2), a nonprofit organization based in Seattle, Washington, founded the ATOM project to highlight the risks of the United States lagging behind in open source software. He says the country needs state-of-the-art open models, in part because relying on foreign models could prove problematic if those models were suddenly phased out or became closed models.
Open models also foster innovation and experimentation between startups and researchers, says Lambert. Additionally, companies with sensitive information need open models that they can run on their own hardware. “Open models are an essential element of AI research, dissemination and innovation, and the United States should play an active role, leading rather than following other collaborators,” says Lambert.
Project ATOM, launched on July 4, makes a compelling case for greater openness and shows how Chinese open-scale models have overtaken American models in recent years.
Ironically, the open source AI movement was started by US social media giant Meta when it released Llama, a frontier open weight model, in July 2023. At the time, the meta saw Llama as a way to break into the AI race. Very quickly, his fresh model became popular among researchers and entrepreneurs.
Since then, Meta and other U.S. AI companies have focused on the idea of developing artificial intelligence at human or superhuman levels, preferably ahead of competitors, resulting in less openness. In recent months, Zuckerberg has rebooted Meta’s artificial intelligence efforts, making a number of high-cost hires and creating a fresh “superintelligence” lab. Zuckerberg also pointed out that Meta may no longer open source its best models.
In turn, China’s technology industry has turned towards greater openness this year. In January 2025, a then-little-known startup DeepSeek released an open source model called DeepSeek-R1, which shocked the world due to its advanced capabilities and the fact that it was trained at a fraction of the cost of mainstream US models. Since then, many Chinese companies have introduced powerful open-scale models with additional innovations.