In the past six months, downloads of Alibaba’s Qwen AI models have exceeded 3 billion, according to a company statement cited by Bloomberg.
The Chinese corporation has made over 460 neural networks publicly available, from which developers have created 300,000 derivatives.
Insights from Hugging Face Data
A day before Bloomberg’s publication, Hugging Face released a semi-annual report on the state of open models. According to their calculations, Qwen was downloaded 2.05 billion times in the first seven months of this year, with 151,448 derivative repositories created. Google’s figure stands at 82,506.
According to Hugging Face, in 2026, Google’s model downloads reached 418 million, while Meta’s were 227 million. However, the platform’s statistics only account for activity within its ecosystem, excluding API requests, private deployments, and other distribution channels.
The report’s authors also emphasized that these figures should not be seen as indicators of market share or actual commercial use.
The number of Qwen derivative repositories grows by 180-210 daily. Of the 28,531 GGUF conversions of these models on the platform, only 54 were created by Alibaba, with the rest by the community.
“Qwen has become part of the standard workflow for developers choosing which model to fine-tune and deploy,” noted Hugging Face.
Why Qwen Surpassed Competitors
Experts attribute Qwen’s position to three factors: regular updates, coverage of all scales—from sub-billion versions to Qwen3.8-Max (2.4 trillion parameters), and the Apache 2.0 license, which does not restrict modifications and commercial use.
The breadth of the model range proved decisive. According to Hugging Face, versions with less than 1 billion parameters account for 83% of all downloads in the platform’s history, while neural networks with over 100 billion account for only 1%.
Labs focused on large LLM lag behind: for instance, Moonshot AI, which rarely releases models smaller than 70 billion, gathered 37 million downloads in a year—about 55 times less than Qwen.
Alibaba’s development also leads in the local deployment segment: its GGUF builds are downloaded 39.6 million times a month, compared to Gemma’s 20.8 million and Llama’s 7.5 million.
Shifts in the Open AI Landscape
The report showed a shift in balance: almost every month this year, the largest open model from China surpassed all U.S. releases in size. Its ceiling ranged from 754 billion to 2.78 trillion, while competitors’ ceilings did not exceed 130 billion in five of the seven months.
In China, 59% of models with more than 20 billion parameters are released under Apache 2.0, 22% under MIT, and none have commercial use restrictions. In the same category, American developers show a different picture: 29% under Apache/MIT, 41% under proprietary terms, and 30% without a specified license.
In the U.S., chip manufacturers AMD and Nvidia, each releasing over 200 repositories, have become leaders in the number of new open models, surpassing AI labs like Google and Meta.
In August, Techno-Nationalism author Alex Capri stated that the United States maintains an advantage in the AI race not so much through individual models but through control over infrastructure.
