Accounting for 41%! China's open-source model downloads surpass those of the United States

```

On another front of the AI arms race, Chinese open-source models are quietly reshaping the competitive landscape with their download numbers.

According to the Hugging Face Spring Report 2026, Chinese open-source models now account for 41% of the platform’s monthly and overall downloads, surpassing US models. Recently, this data was again highlighted by Hugging Face CEO Clem Delangue, sparking broad discussion on social media.

Hugging Face is currently the world’s largest open-source AI model hosting platform, hosting nearly 3 million public models and 1 million public datasets. On average, a new code repository is created every seven seconds. Half of Fortune 500 companies have already deployed their own private or open-source models on this platform.

Moreover, the six most popular models on OpenRouter all come from Chinese institutions such as Tencent, Xiaomi, DeepSeek, MiniMax, and Zhipu.

Hugging Face CEO Clem Delangue (right)

Chinese Models Dominate the Top Six on OpenRouter

Beyond download data, the changes on the usage side are equally striking.

On the AI model invocation platform OpenRouter, the current six most popular models all come from Chinese institutions—Tencent, Xiaomi, DeepSeek, MiniMax, and Zhipu are all included. Anthropic’s Claude Opus 4.7 ranks seventh.

Zhipu recently released the open-source model GLM-5.2, aimed at agentic coding tasks. In relevant benchmark tests, it stands on the same competitive tier as Anthropic's latest model.

Infrastructure data corroborates this trend as well. According to Vercel platform data, in June 2026, open-source models handled nearly one-third of AI requests on the platform, undertaking many infrastructure-intensive workloads, while closed-source models are gradually retreating to the high-cost premium tier.

Enterprise Shift: From “Rental” to “Ownership”

The core logic driving this shift is cost and control.

Delangue said: "If you are an AI company or a tech company, you don’t want to outsource your core capabilities to a black-box API that you cannot control." He pointed out that after seeing the bills for scaling frontier closed-source models, companies are reassessing the value of owning their own AI.

Microsoft CEO Satya Nadella has issued similar warnings against dependency on a single vendor. He said: "If learning only flows in one direction, economic value will concentrate in the hands of those who own the learning infrastructure, rather than flowing to creators of knowledge." He advocates for dispersing learning infrastructure among all enterprises.

UBS Securities analyst Karl Keirstead’s team noted in their AI research report dated June 23, 2026, that about 60% of enterprises have put restrictions on AI spending, with the main action being the introduction of limits on Token usage.

“Model Routing” Becomes Mainstream: Premium Models Shift from Default to Luxury

Enterprise response strategies are systematically reshaping the benefit landscape of the AI industry chain.

According to UBS, “model routing” is now the core technical action for Token optimization—assigning different tasks to different models, calling the most expensive models only for complex reasoning, critical code, and long-context analysis. Simple tasks are handled by low-cost or even Chinese open-source models.

Price differences are the direct driver. Take Anthropic as an example: Haiku 4.5 output costs $5 per million Tokens, Opus 4.5–4.8 costs $25, and the highest-end models reach $50—token output prices differ tenfold from low-end to high-end.

It is against this backdrop that Chinese open-source models are entering the corporate procurement spotlight. According to a case cited in the UBS report, a major global bank has already deployed Alibaba's Qwen locally to balance the usage costs of premium models like Claude. MiniMax, Kimi, Qwen, DeepSeek, and GLM are listed in AWS Bedrock's model menu; Microsoft Azure AI Foundry also provides DeepSeek integration.

Can Open Source Decide the Direction of the AI Race?

Delangue’s assessment is more direct. According to PodcastAlpha, Delangue believes open source is an accelerator of AI leadership, “If China continues to lead open source, it may take the lead in AI overall within one or two years.”

Anthropic CEO Dario Amodei takes a different stance. He believes that releasing increasingly powerful open-source models carries risks because once released, they are hard to control.

Delangue rebuts: “Locking it behind the doors of a few players doesn’t make it safe.” He adds, transparency allows defenders “to patch cybersecurity risks already exposed by open-source models.”

Risk Disclaimer and Exemption ClauseThe market is risky, and investment requires caution. This article does not constitute personal investment advice, nor does it take into account the individual user’s specific investment goals, financial situation, or needs. Users should consider whether opinions, views, or conclusions in this article fit their specific circumstances. Investing based on this is at your own risk. ```