According to the latest analysis by the Human-Centered AI Institute at Stanford University, multiple Chinese institutions have now launched the most advanced artificial intelligence models and have surpassed their American counterparts in global distribution and application. Early in 2025, with the release of the R1 model by Deepseek, the global spotlight was on this Chinese startup. However, researchers point out that China's open-weight AI ecosystem is much larger and more complex than people might imagine.
The study shows that Alibaba's Qwen model family became the most downloaded language model family on the Hugging Face platform in September 2025, surpassing Meta's Llama. Between August 2024 and August 2025, downloads from Chinese developers accounted for 17.1%, slightly higher than the 15.8% from the United States. More notably, 63% of new fine-tuned models are based on Chinese foundational models.
Stanford researchers emphasized that Deepseek is not the only significant player. In addition to this Hangzhou-based startup, multiple Chinese organizations have publicly released high-performance models. These organizations include well-known tech giants such as Alibaba, Tencent, Baidu, Huawei, and ByteDance.
Due to U.S. restrictions on exports of the most powerful AI chips since October 2022, Chinese developers have gradually turned toward more efficient open-weight models. At the same time, the licensing terms for these models have also become more relaxed, allowing near-unlimited use and modification.
Globally, an increasing number of countries and companies are beginning to adopt Chinese open-weight models. Singapore's national AI program is building its flagship model based on Alibaba's Qwen, while the U.S. company Meta has acquired a startup that uses a Chinese open-weight model.
However, researchers also point out that security issues cannot be ignored. U.S. government tests showed that the Deepseek model is twelve times more vulnerable to attacks than similar U.S. models. Although the Chinese government's support for open AI development remains uncertain, it has consistently advocated for technological equity on the international stage.
Key Points:
1. 📈 Chinese open-weight AI models have surpassed American ones in global distribution and application, with Alibaba's Qwen becoming the most popular language model.
2. 🌍 Multiple Chinese organizations are involved in AI model development, including well-known technology companies and emerging startups.
3. 🔒 Despite the global adoption of Chinese AI models, security issues require attention, as some models show significant vulnerability to attacks.
