Alibaba's open-source Qwen models have surpassed the 3 billion download milestone.
Chinese tech giant Alibaba has reached an impressive milestone: over the past six months, the total number of downloads of its open-source AI models from the Qwen family has exceeded 3 billion. This landmark event underscores the rapidly growing influence of Chinese developments on the global artificial intelligence market.
To date, the corporation has provided the community with more than 460 neural networks in open access. Developers worldwide are actively using this foundation, creating over 300,000 derivative models based on it, which indicates the formation of a powerful ecosystem around Qwen.
Analysis of the Hugging Face Ecosystem
My data, obtained through an analysis of the Hugging Face platform over the past seven months, fully correlates with these figures. According to monitoring, Qwen was downloaded 2.05 billion times, and the number of derivative repositories reached 151,448. For comparison, Google's figure is only 82,506. Notably, for the entire year of 2026, Google model downloads are estimated at 418 million, while Meta's stand at 227 million.
It is important to understand that Hugging Face statistics only account for activity within its ecosystem and do not include API requests or private corporate deployments. Nevertheless, even with these limitations, Qwen's growth dynamics are impressive: the number of derivative repositories increases daily by 180–210. It is also telling that out of 28,531 GGUF conversions of these models, only 54 were created by Alibaba itself—the rest is generated by the community.
Success Factors and the Balance of Power
Qwen's success is no accident. I see three key factors that ensured its leadership: regular updates, coverage of all scales—from compact versions with sub-billion parameters to the giant Qwen3.8-Max (2.4 trillion parameters)—and, critically, the Apache 2.0 license, which removes restrictions on commercial use.
The breadth of the model lineup proved decisive. Analysis shows that 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%. This explains why laboratories focused on giant LLMs, such as Moonshot AI, are lagging behind, gathering only 37 million downloads over the year—roughly 55 times less than Qwen. In the local deployment segment, the gap is also significant: Qwen GGUF builds are downloaded 39.6 million times per month, compared to 20.8 million for Gemma and 7.5 million for Llama.
The report also revealed a shift in the global balance: almost every month, the largest open-source model from China surpassed all American releases in size. In China, 59% of models with more than 20 billion parameters are released under Apache 2.0, while in the US, 41% of large models use proprietary terms. Notably, in the States, chip manufacturers AMD and Nvidia, rather than traditional AI laboratories, became leaders in the number of new open models.
My comment: The three-billion mark is not just a number but a marker of a paradigm shift. Qwen has become the de facto standard for developers who value flexibility and openness. The US retains control over infrastructure, but in the open-weights segment, China has undoubtedly seized the initiative, and this gap will only widen.