Alibaba's Qwen family of open-weight models surpassed 3 billion downloads on Hugging Face in the six months to mid-August 2026, compared to 418 million for Google and 227 million for Meta in the same period, according to a Hugging Face platform report. Over 300,000 derivative models have been built on Qwen, and the Hugging Face report characterizes it as embedded in developers' default fine-tuning workflows. Separately, Chinese labs released open models with parameter counts ranging from 754 billion to 2.78 trillion monthly in 2026, while US labs stayed below 130 billion parameters in five of seven months covered. For financial institutions, the data points to Chinese open-weight models gaining structural position in the global developer stack, which bears on how rent and switching costs distribute across the AI value chain independently of closed-model competition. Also reported by: https://www.globaltimes.cn/page/202608/1368313.shtml
437 other tracked developments touching the same claims.