BEIJING, Nov 7 (APP):From an AI-powered sewing machine that monitors operations and makes real-time adjustments, to a device-testing robotic arm that boosts efficiency by 75%, and even a crab-shaped robot that cleans shoes, more than 1,000 innovations are on display at the Light of Internet Expo, held alongside the World Internet Conference Wuzhen Summit in Zhejiang Province. Together, these products offer a glimpse of how intelligent technologies are reshaping everyday …
China’s open-source AI ecosystem fuels global innovation
BEIJING, Nov 7 (APP):From an AI-powered sewing machine that monitors operations and makes real-time adjustments, to a device-testing robotic arm that boosts efficiency by 75%, and even a crab-shaped robot that cleans shoes, more than 1,000 innovations are on display at the Light of Internet Expo, held alongside the World Internet Conference Wuzhen Summit in Zhejiang Province.
Together, these products offer a glimpse of how intelligent technologies are reshaping everyday life. Behind many of them lies a growing force: China’s open-source AI ecosystem.
At the summit, China’s first open-source-themed ecosystem experience center was launched. ModelScope, China’s largest open-source AI model community, brings together more than 120,000 open-source models and over 5,500 MCP services. Nearly 23,000 AI applications are brewed here, around 95% developed by individual developers.
“ModelScope now serves more than 33 million developers from 239 countries and regions,” said Joe Tsai, co-founder and chairman of Alibaba Group, which founded and operates the platform.
Speaking at the opening ceremony of the 2026 summit on November 3, Tsai said Alibaba had open-sourced more than 470 Qwen models.
As AI applications proliferate and demand for computing power continues to outstrip supply, Chinese technology companies are stepping up efforts to open-source their models and make their ecosystems more accessible, he added.
With costs lowered to as little as RMB 1 per million tokens, the empowering impact of China’s open-source models is increasingly visible beyond its borders.
According to a report by Hugging Face, the world’s largest AI open-source community, models developed in China accounted for 41% of downloads on the platform over the past year. Meanwhile, the cumulative number of downloads of open-source models from several Chinese technology companies across major global open-source communities has exceeded 10 billion.
“By lowering the barriers to access, it gives researchers, students and developers with limited resources more opportunities to participate. It can also help address needs involving less widely spoken languages, specialized applications and adaptation to specific hardware, while reducing the risk of market monopolies,” said Liu Dayiheng, head of the Qwen large language model project at Alibaba’s ATH business group.
For medium- and small-sized countries and companies, this implies an unprecedented, affordable opportunity to catch up with the trend.
Sandro Tsirekidze, chairman of Georgia’s Digital Governance Agency, told China Economic Net that Georgia signed a cooperation agreement with China in July to use Chinese open-source large language models (LLMs) as the technological foundation for developing comprehensive AI agent solutions tailored to specific needs and projects.
“Demand for computing power outstrips supply. In Uganda, a nonprofit organization used open-source Qwen models to develop models for local languages, covering 68 African languages and helping more people overcome language barriers and use AI,” Tsai shared.
Bernd Holznagel, a professor at the University of Münster in Germany, said 40% to 50% of small and medium-sized enterprises in Germany are now using large language models developed in China.
The pace of open-source development in China is still gathering.
Just this week, Tencent unveiled its Hunyuan Hy4 model and announced plans to open-source its weights in the near future on November 4th. On the 5th, Baidu released the base weights of ERNIE 5.5 Flash Base, while Moonshot AI made the full weights of Kimi K3.1 Pro available on the same day. On the 6th, Ant Group introduced Ling-3.2 Flash, part of its Ling family of models.
“Open source means that ‘last-mile’ applications – minority languages and niche vertical domains – will actually have people working on them,” Liu said.
At the Wuzhen Summit’s main forum, an action initiative for jointly building global open AI was released. It sets out 18 proposed actions and calls on countries to jointly develop and share computing-power pools and high-quality datasets, establish seed funds for open-source projects, and support countries in the Global South in developing localized applications on open-source foundations.
