The Chinese corporation Alibaba has announced that the number of downloads of its open Qwen family models over the past six months has exceeded three billion. The company has made over 460 neural networks freely available, on the basis of which developers have created 300,000 derivative versions. Shortly before this, the Hugging Face platform published a half‑year report, according to which Qwen models were downloaded 2.05 billion times in the first seven months of this year, and the number of derivative repositories reached 151,448. For comparison, Google’s figure was 82,506 repositories. At the same time, in 2026, Google’s model downloads reached 418 million, while Meta’s reached 227 million. The platform’s statistics take into account only activity within its ecosystem and do not reflect actual commercial use.

Experts attribute Qwen’s leadership to three key factors: regular updates, coverage of all model scales, and the free Apache 2.0 license. The diversity of the model range turned out to be decisive: versions with fewer than one billion parameters account for 83 % of all downloads over the platform’s history, while neural networks with over 100 billion parameters account for only 1 %. Labs focused on large models are noticeably lagging behind: Moonshot AI accumulated 37 million downloads in a year, which is about 55 times less than Qwen’s figure. Alibaba’s development also leads in the local deployment segment: its GGUF builds are downloaded 39.6 million times per month, while Gemma is downloaded 20.8 million times, and Llama — 7.5 million times.

The Hugging Face report also recorded a shift in the balance of power in open‑source artificial intelligence towards China. Almost every month of the current year, the largest open model from China surpassed all releases from the United States in terms of size: its parameter ceiling ranged from seven hundred fifty‑four billion to two trillion seven hundred eighty billion parameters, whereas for American competitors it did not exceed one hundred thirty billion in five out of seven months. In China, fifty‑nine percent of models with more than twenty billion parameters are released under the Apache 2.0 license, and twenty‑two percent under the MIT license, and none have restrictions on commercial use. In the US, the picture is different: twenty‑nine percent are under Apache or MIT, forty‑one percent are under proprietary terms, and thirty percent are without a specified license. It is noteworthy that the leaders in terms of the number of new open models in the US are chip manufacturers AMD and Nvidia, each of which has released more than two hundred repositories. Earlier, Alex Capri, the author of the book Techno‑Nationalism, stated that the United States maintains an advantage in the AI race thanks to its control over infrastructure.