At the 2026 World Internet Conference Wuzhen Summit held in Wuzhen, Zhejiang, the topic of the AI open-source ecosystem became the focus of everyone’s attention. In recent years, China’s large AI models have continued to win fans around the world thanks to continuous technological iteration, high performance, and strong cost-effectiveness, showing powerful international competitiveness.

Global data confirms: Chinese large models lead in usage, with DeepSeek at the top

The popularity of Chinese large AI models is backed by detailed data. According to statistics from OpenRouter, a global AI model aggregation platform, during the week from October 26 to November 1, the token usage of Chinese large models exceeded that of the United States for 27 consecutive weeks. At the same time, among the top five models by total weekly token usage globally, Chinese large AI models took four spots, with DeepSeek ranking first.

Gao Xi, a research engineer at DeepSeek, noted that from early basic chat and translation functions to today’s deep logical reasoning and ability to truly solve users’ real-life problems, Chinese large AI models have developed at an unimaginable pace, maintaining a rhythm of significant technological innovation and breakthroughs almost every year.

International experts give thumbs up: Silicon Valley and European SMEs become loyal users

In terms of international applications, Chinese open-source models have also gained broad recognition. John Hoffman, Chairman of the GSMA, said that some of China’s large language models and the small language models under development have made tremendous progress, and the vast majority are built on open-source platforms.

Bernd Holznagel, a professor at the University of Münster in Germany and a hands-on user of DeepSeek, observed that many companies in Silicon Valley in the United States and nearly half of small and medium-sized enterprises in Germany now widely use Chinese open-source large AI models in their daily work.

Joseph Tsai, co-founder and chairman of Alibaba Group, pointed out that Chinese open-source models have strong overall performance and high inference efficiency, giving them very strong competitiveness globally. Over the past year, global AI application scenarios have been rapidly expanding into fields such as industrial manufacturing, healthcare, public services, corporate office work, and technology R&D. Chinese tech companies’ continued efforts to increase open-source and ecosystem openness have kept key indicators such as open-source model downloads and the number of derivative models in the lead.

Breaking down “small yards with high fences”: open-source ecosystems help close the digital divide

Experts at the event generally agreed that if AI technology is locked tightly behind “small yards with high fences,” sparks of innovation will find it hard to spread, ultimately hindering the overall progress of global technology. The open and shared “open-source” trend, however, is opening up broader imaginative space for future AI innovation.

Tsai said that open source can greatly reduce R&D costs for SMEs, especially enabling developers in emerging economies to quickly create AI applications that meet their countries’ needs based on open-source models. For example, a nonprofit organization in Uganda developed a local language model based on the Qwen model. It now covers 68 African languages, helping more local people break through language barriers and successfully access and use advanced AI technology.

Liu Dayiheng, head of the Qwen large language model project at Alibaba’s ATH business group, said bluntly that without open source, the core technology of large AI models might remain monopolized by a small number of people and companies for a long time. Through open source, scholars and students lacking resources can acquire cutting-edge knowledge and participate in technological innovation at extremely low cost; and minority languages, special vertical applications, or specific hardware compatibility issues that big tech companies may not have the time to address can also be effectively solved with the collective wisdom of open-source communities. In addition, the high cost-effectiveness of open-source models themselves also plays a positive role in preventing price monopolies in the market.