Alibaba’s Qwen family surpasses 3 billion downloads as Chinese open-source AI models outpace US alternatives across global developer metrics.
Alibaba’s open source Qwen family surpassed 3 billion cumulative downloads over six months, overtaking Meta and Google on Hugging Face as of 14 August 2026. In comparison, Google recorded 418 million downloads and Meta reached 227 million.
Alibaba has released over 460 Qwen models, spawning 300,000 derivative creations. Meanwhile, OpenRouter data for 3–9 August showed global AI token usage hitting 69 trillion, with Chinese models claiming 34.25 trillion tokens to outpace US models for the 15th straight week.
Chinese releases consistently surpassed US parameter scales throughout 2026, reaching ceilings between 754 billion and 2.78 trillion parameters. Moonshot AI launched Kimi K3, the world’s largest open model at 2.8 trillion parameters. Zhipu AI introduced GLM-5.3 to democratise security defence tools, while DeepSeek launched Harness, an agent runtime framework for automated code testing and fixing.
Supporting 119 languages and regional dialects, Qwen is rapidly expanding enterprise deployments across Southeast Asia and Africa. In response to potential regulatory barriers, major US technology companies, including NVIDIA, Microsoft, IBM, Meta, and OpenAI, issued a joint statement urging domestic policymakers to safeguard open-weight models. Concurrently, nearly 200 Silicon Valley startups under the Little Tech Association submitted a formal letter opposing US government restrictions on utilizing Chinese open-weight AI models.
