Open-source AI models from China now dominate developer downloads. Does the frontier still matter?
Chinese open-weight models took 41% of Hugging Face downloads this spring. Developers are building with cheaper, customisable alternatives, and some big names think that shift is permanent.
Key points
- Chinese open-weight models, meaning AI models whose underlying code is publicly released for anyone to use, accounted for 41% of all downloads on the Hugging Face developer platform in spring 2025.
- On OpenRouter, a service that lets developers access many AI models through one connection, the six most-downloaded models all come from Chinese firms including Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai.
- Open models handled nearly one third of all AI requests on Vercel, a major app-hosting platform, in June 2025.
- Half of all Fortune 500 companies now use Hugging Face to deploy their own private or open-source AI models, according to Hugging Face CEO Clem Delangue.
- Beijing-based Z.ai recently released GLM-5.2, an open-weight model that competes directly with Anthropic's latest models on finding software security flaws.
For weeks this summer, the tech press fixated on Anthropic's newest models and Washington's fight over who should be allowed to use them. While that debate raged, developers quietly made a different choice.
They turned to open-weight models, AI models whose internal code and parameters are published openly so anyone can download, modify, and run them, often for free, on their own computers. No subscription. No API bill. No terms of service that can change overnight.
The numbers are striking. Chinese open-weight models took 41% of downloads on Hugging Face, the largest public platform for sharing AI models, this spring. On OpenRouter, the top six spots all belong to models from Chinese labs. Anthropic's Claude Opus 4.7 sits seventh, as first reported by TechCrunch AI.
Hugging Face's Delangue put the business logic plainly: "If you're an AI company or a technology company, you don't want to outsource your core capabilities to another company, to a black box API that you don't control, don't have any visibility on."
A new model repository, essentially a storage folder for an AI project, is created on Hugging Face every seven seconds. The platform now hosts almost three million public models.
What does this mean for ordinary businesses?
For any company paying monthly fees to use ChatGPT or Claude through an API, an application programming interface, meaning a paid connection to a company's AI service, the message is simple: cheaper, controllable alternatives exist and large companies are already using them. Half of Fortune 500 firms are on Hugging Face, Delangue says.
Microsoft CEO Satya Nadella made a similar point recently, warning companies not to tie themselves to a single AI provider. He argued that when learning and data flow only toward the model owner, economic value follows, leaving customers with little to show for their spending.
Not everyone welcomes this shift. Anthropic CEO Dario Amodei has argued that releasing powerful model weights openly is dangerous because, once published, those models cannot be recalled or updated if problems emerge. Others worry bad actors could use open models to spread disinformation or assist in cyber attacks.
Delangue disagrees. "The biggest risk in AI is concentration of power," he said. Transparency, he argues, lets security teams spot and fix problems faster than secrecy does.
For now, frontier models from Anthropic and OpenAI still handle complex, high-value tasks. But the direction of travel is clear: for everyday production work, open and cheap is winning.
