Alibaba’s AI models have strengthened China’s position in the race to become the foundation developers build AI products on.
The company said its Qwen family has surpassed 3 billion cumulative downloads across platforms, putting it ahead of open AI offerings from Meta, Google and other rivals. Bloomberg reported the figure based on an emailed statement from Alibaba.
The company has released more than 460 Qwen models as open source, which have generated more than 300,000 derivative models, Alibaba said.
Hugging Face’s Aug. 14 State of Open Models report provides a narrower but independently measured view. It recorded about 2.045 billion Qwen downloads on its platform during the first seven months of 2026, compared with 418 million for Google and 227 million for Meta. The Hugging Face figures exclude other distribution platforms, including China’s ModelScope.
Alibaba’s 3 billion figure covers downloads across platforms, while Hugging Face measures activity on its own hub.
Developers are building on Qwen
Downloads are important for open models because developers can take them, modify them, fine-tune them for specific uses and incorporate them into new applications. Qwen’s influence extends beyond downloads. Hugging Face counted 151,448 models derived from Qwen on its platform, 2.6x the number derived from Meta models.
“Qwen has become part of the default workflow for developers deciding what models to fine-tune and deploy,” Hugging Face said in its report.
Alibaba is also distributing Qwen through its cloud business to enterprise customers in markets including Southeast Asia and Africa, Bloomberg reported. That gives the models another route into commercial deployments.
The open-model race is changing
The rise of Qwen comes as Chinese AI companies increasingly compete with U.S. developers on open models. Moonshot AI, DeepSeek and other Chinese labs have released large models designed to offer strong performance while remaining relatively easy to adapt.
Hugging Face also found that leading Chinese AI laboratories generally released larger open models than their U.S. counterparts in 2026. But parameter count alone does not establish that a model is more capable or efficient, particularly with mixture-of-experts architectures.
The licensing strategy is also significant. Of 178 Chinese releases above 20 billion parameters tracked by Hugging Face in 2026, 59% used the Apache 2.0 license and 22% used the MIT license. Almost none carried noncommercial restrictions, although Hugging Face noted that some recent large releases have introduced tighter terms.
The catch: Downloads aren’t everything
Qwen’s numbers show growing developer interest, but downloads are only one measure of influence. They do not capture private deployments, API usage or activity on platforms outside the dataset being measured.
That makes the more important signal the ecosystem forming around Qwen. If developers continue turning Alibaba’s models into thousands of specialized systems, Qwen could gain influence even without directly powering every application.
For Alibaba, that could translate into something more valuable than a download record: a larger developer base feeding demand for its AI and cloud services. American labs are responding. Meta and Nvidia have both released new open models in recent weeks, and Nvidia has been explicit that it's chasing China on open weights.
Read more: Alibaba’s download momentum coincides with the release of Qwen3.8-27B under the Apache 2.0 license, showing how the company is using accessible models to expand its developer ecosystem.