Hugging Face says developers made 151K+ derivatives based on Qwen models, topping others, making Qwen one of the largest foundations in the open model ecosystem (Hugging Face)

Developers have created over 151,000 derivatives based on the Qwen models, solidifying Qwen's status as one of the largest foundations in the open model ecosystem, according to Hugging Face. In a report covering January to August 2026, the distribution of models and datasets on the Hugging Face hub has grown significantly, with public model repositories increasing from 2.43 million to 2.96 million and datasets rising from 711,000 to 1 million.

The majority of models have under 200 lifetime downloads, and a mere 1.5% of repositories account for 99.2% of all downloads. In 2026, Chinese labs have bypassed the traditional progression of releasing smaller models and rapidly scaling up to larger ones. In each month, the most performant open model from a Chinese lab outperformed anything an American lab released.

Chinese labs like Moonshot, MiniMax, Xiaomi, and Z.ai release models that are too large for developers to run on their own hardware, while Tencent and Alibaba's Qwen span the entire range of model sizes. The U.S. hardware vendors NVIDIA and AMD have recognized that open models can help sell chips, with both companies releasing more than 200 new model repositories this year.

However, the center of gravity in AI development has shifted, with Google and Meta now ranking below NVIDIA in new model releases, despite their historical role in open model publishing. Most U.S. releases above 100B parameters this year are built on top of Chinese models, with only a few original American models at this scale. Chinese open models are increasingly optimized for domestic chips, creating a distribution and optimization layer rather than model creation.