No. 1 in the world: Alibaba’s Tongyi Qianwen AI model has been downloaded more than 3 billion globally, crushing Meta and Google

IT House reported on August 15 that Alibaba’s open source AI model Qianwen Family has been downloaded more than 3 billion times globally in the past six months, surpassing competitors such as Meta and Alphabet, and becoming the world’s most downloaded AI model. According to the "State of Open Models" report released by the open source AI community Hugging Face on August 14, local time, in 2026, Google models will be downloaded approximately 418 million times and Meta approximately 227 million times. Alibaba said that the Qianwen series has open sourced more than 460 models and derived more than 300,000 derivative models. The Hugging Face report states that Qwen's download data makes it "one of the largest foundations in the open AI ecosystem" and "Qwen has become part of the default workflow for developers to decide which models to fine-tune and deploy." The report shows that the number of derivative models based on Qwen reached 151,448 on Hugging Face, which is 2.6 times that of Meta and 4.7 times that of the Llama series warehouses. The number of Google derived models is 82,506. The Hugging Face report shows that the largest open model parameter scale released by China’s cutting-edge AI laboratories in 2026 generally exceeds that of American laboratories. Chinese lab monthly caps range from 754 billion to 2.78 trillion parameters, and the U.S. falls below 130 billion parameters in most months. Licensing for Chinese models is also more relaxed. Among the 178 Chinese models with more than 20B parameters released in 2026, 59% adopt the Apache 2.0 protocol, 22% adopt the MIT protocol, and none have non-commercial restrictions. The report pointed out that parametric models below 1B accounted for 83% of the historical cumulative downloads, and those above 100B accounted for only 1%. Qwen’s local deployment model in GGUF format has been downloaded 39.6 million times per month, close to G