Qwen from Alibaba has surpassed the 3 billion download mark: how Chinese models are capturing the world of open AI
Chinese tech giant Alibaba has reached an impressive milestone: over the past six months, the total number of downloads of their open-source AI models from the Qwen family has exceeded 3 billion. This is a colossal figure that propels the Chinese corporation to absolute leadership in popularity among developers worldwide.
To date, Alibaba has released more than 460 different neural networks to the public, and the number of derivative models built on their basis has reached 300,000. This indicates the formation of a powerful developer ecosystem around Qwen, with developers actively using these models as a foundation for their own projects.
Hugging Face Data: Impressive Numbers
Analysis of the Hugging Face platform, which is the main hub for open-source AI, confirms this trend. Over seven months of the current year, Qwen models were downloaded 2.05 billion times, and the number of their derivative repositories reached 151,448. For comparison, Google's figure is only 82,506. For the entire 2026, Google model downloads are estimated at 418 million, and Meta's at 227 million. It is important to understand that Hugging Face statistics only account for activity within its own ecosystem, excluding API requests and private deployments, making the real scale of Qwen's distribution even more significant.
Notably, 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 itself—the rest were made by the community. This is a clear indicator that Qwen has become the de facto standard in the workflow of developers choosing which model to fine-tune and deploy.
The Secret to Success: Strategy, Not Coincidence
Qwen's superiority over competitors is not accidental. I see three key factors that ensured this success:
First, regular updates to the model lineup. Second, incredible breadth of coverage—from compact versions with sub-billion parameters to the giant Qwen3.8-Max with 2.4 trillion parameters. Third, and most importantly, the Apache 2.0 license, which imposes no restrictions on modification and commercial use.
It is precisely the breadth of the lineup that proved decisive. According to Hugging Face data, versions with fewer than 1 billion parameters account for 83% of all downloads in the platform's history, while neural networks over 100 billion account for only 1%. Competitors focused on massive LLMs, such as Moonshot AI, which release virtually no models under 70 billion, gathered only 37 million downloads over the year—about 55 times less than Qwen. In the local deployment segment, Qwen GGUF builds are downloaded 39.6 million times per month, leaving Gemma (20.8 million) and Llama (7.5 million) behind.
Shifting Balance of Power in Open AI
The report also demonstrates a fundamental shift in the balance of power. Almost every month this year, the largest open model from China surpassed all U.S. releases in size. The parameter ceiling for Chinese models ranged from 754 billion to 2.78 trillion, while for American competitors it did not exceed 130 billion in five of seven months.
Licensing policy also plays into China's hands. There, 59% of models with more than 20 billion parameters are released under Apache 2.0 and 22% under MIT, with no restrictions on commercial use. Among American developers in the same category, only 29% are under open licenses, 41% under proprietary terms, and 30% without any license specified. Notably, in the U.S., the leaders in the number of new open models were not AI labs but chip manufacturers AMD and Nvidia, which each released more than 200 repositories.
My analysis: Qwen's success is not just a marketing victory but the result of a sound ecosystem strategy. Alibaba bet on accessibility and flexibility, which, in an environment where developers increasingly seek efficient solutions for local deployment, proved far more important than the race for record parameter counts. This is a serious signal for Western labs that still rely on closed and ultra-large models.