Hugging Face: Open Models Redefine the Future of AI

Hugging Face: Open Models Redefine the Future of AI

For several weeks this summer, the AI industry was focused on the latest cutting-edge models from Anthropic and Washington's struggle to control access to these models. But while everyone was watching this situation, developers continued to build—without waiting for permission from companies like Anthropic and OpenAI.

Chinese open-weight models accounted for 41% of downloads on Hugging Face this spring, surpassing American models. On OpenRouter, the six most popular models are all open models from Chinese companies, including Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. Anthropic's Claude Opus 4.7 ranks seventh at the time of writing. Furthermore, data from Vercel shows that open-weight models handle a significant portion of the heavy infrastructure for AI applications, while closed models operate as a more expensive premium layer. In June, open models processed nearly one-third of AI requests on the platform.

These platforms capture only a part of the AI ecosystem; in particular, they omit sessions hosted by major labs, which likely represent the majority of usage for OpenAI and Anthropic. But the growing market share of open-source models raises a difficult question: what is the relevance of cutting-edge models if most production AIs operate on cheaper, customizable alternatives?

Some see the growth of open-source models as a sign that the smartest models may ultimately be used only for very specialized use cases. “Perhaps in a few years, cutting-edge models will be reserved for experimentation and certain high-value tasks, and most production workloads will actually be powered either by private models within companies or by open-source models,” said Clem Delangue, CEO of Hugging Face, during a recent episode of Equity.

Hugging Face is a platform and community of developers best known for hosting, sharing, and helping companies deploy open models. Delangue claims that clients and members of the Hugging Face community increasingly tout the benefits of owning their own AI models rather than renting them, a trend that has gained momentum after becoming aware of the costs associated with scaling closed cutting-edge models.

“If you are an AI company or a tech company, you don’t want to outsource your core capabilities to another company, to a black-box API that you don’t control, that you have no visibility into, and that you don’t really own a stake in,” Delangue said.

This shift, Delangue argues, is reflected in the activity on Hugging Face. A new repository is created every seven seconds on the platform, which hosts nearly three million public models and one million public datasets, according to Delangue. This indicates a different reality from the “one model to rule them all,” he says. In reality, it looks more like companies using many different models, many of which are customized for their specific use cases. Half of all Fortune 500 companies use Hugging Face to deploy their own private models and open-source models, he claims.

The growing popularity of open models coincides with a steady stream of increasingly powerful releases from Chinese AI labs.

Every few months, another Chinese AI company releases a powerful open-weight model that is cheaper to deploy and easier to customize than closed competitors, undermining the economics of proprietary AI in which American companies have invested billions. Most recently, the Beijing-based AI company Z.ai launched an open-weight model called GLM-5.2, which excels in agentic coding and competes with the latest models from Anthropic in identifying security vulnerabilities.

Delangue is not the only executive advocating that companies should avoid tying themselves to a single model provider.

Microsoft CEO Satya Nadella recently warned against vendor lock-in, arguing that data control should be a major concern for companies using AI.

“While the great innovation that comes from model providers having fair usage rights to train models on public data is necessary, I find it ironic that the status quo is then to turn around and impose restrictive conditions on distillation, and reserve the right to learn from customer usage and interaction data,” Nadella said. “If learning only flows in one direction, the economic value converges to the owners of the learning infrastructure rather than to the creators of the knowledge itself. Therefore, it is imperative that we distribute the learning infrastructure to every company so they can control their own learning loop.”

The rise of open models has also intensified the debate over whether increasingly powerful models should be widely accessible.

Anthropic CEO Dario Amodei has argued that the scale of powerful open-weight models could become dangerous because, once released, they become difficult to control. Others have contended that open models are more easily accessible to malicious actors who could use them to spread misinformation or conduct cyberattacks or biological warfare.

Delangue sees the trade-off differently.

“The biggest risk in AI is the concentration of power,” Delangue said. “The way to make the world safer, in my opinion, is to level the playing fields and create transparency around these models.”

Transparency means that advocates can more easily “correct cybersecurity risks they already know open-source models can exploit,” he said.

The Hugging Face leader argues that keeping powerful models closed does not eliminate the risks associated with advanced AI systems, partly because it is easy to circumvent the safeguards of cutting-edge model APIs and steal the weights to disseminate them openly. Restricting powerful models, Delangue argues, simply concentrates technology in the hands of a few companies while reducing transparency about how systems operate.

“You don’t make things safer by keeping them behind closed doors for a few players,” Delangue said. “You make them more dangerous because you create an asymmetry of power and an asymmetry of capabilities.”