Qwen's reach is Alibaba's case for open models, but its reported plan to charge large commercial users of an upcoming open-weight Qwen makes license terms the revenue question.
Alibaba Group chairman Joe Tsai used the opening ceremony of the 2026 World Internet Conference Wuzhen Summit on November 3 to argue that open source is the only route to a sustainable AI industry. Only open source can make AI widespread, he said, and only widespread use can create a "sustainable, inclusive" intelligent era, according to a report by The Paper (澎湃新闻) republished by Sina. He backed the case with numbers. Alibaba’s Qwen has open-sourced more than 470 models, which have been downloaded more than 4 billion times and used to build more than 330,000 derivative models. Tsai gave those figures in his speech, AASTOCKS reported.
Tsai framed open source as an answer to scarcity. Demand for AI keeps rising while the supply of computing power falls short, he said, and open models cut research costs for small and medium-sized companies. He described that as especially valuable for developing countries and for languages with few digital resources, citing a nonprofit in Uganda that built a model covering 68 African languages on Qwen.
The argument itself is familiar. Tsai made a similar case to a European audience last month, as EastFrontier reported in October. The setting this time carried more official weight: the Wuzhen Summit is backed by the Cyberspace Administration of China and the Zhejiang provincial government, Dataconomy noted. This year’s edition made open source its headline theme for the first time in the event’s 13 years, the 21st Century Business Herald reported.
Alibaba’s figures put Qwen at the center of China’s open-model push
Tsai said Chinese open models lead the world on measures such as downloads and the number of derivative models, and described Qwen as the world’s leading open-source model family. The 21st Century Business Herald put Alibaba’s count at more than 460 models released over three-plus years, slightly below the figure cited from the speech.
The distribution channel matters as much as the models. ModelScope (魔搭), the open-source community Alibaba created and runs, has 33 million users across 239 countries and regions and hosts 280,000 models, 40,000 datasets and 12,000 MCP servers, the connectors that let AI agents use outside tools, according to The Paper. Tsai said ModelScope’s role is to be based in China and serve the world. At the summit’s exhibition, Alibaba showed its T-Head chips, Alibaba Cloud’s AI services, Qwen models and AI applications.
Qwen’s own team spelled out what openness costs
Alibaba’s speakers also named the limits of the approach they champion. Liu Dayiheng (刘大一恒), who leads Qwen’s large language model project, said open source lets developers fill niche, industry-specific and hardware-specific gaps that top AI companies may ignore, and that open models keep the market’s models good value for money and stop prices from being monopolized, the 21st Century Business Herald reported.
He also named two weaknesses. Open models cannot be recalled once released, and they lack external guardrails, so they depend on safety built into the model, which can be bypassed. Model restrictions alone cannot solve governance, he said, which also needs laws, industry norms and supporting infrastructure. Gao Xi (高熙), a research engineer at DeepSeek, argued at the summit that publishing methods draws capable peers into the work and speeds up progress. John Higgins, chair of the International AI Governance Association, called for a unified evaluation system and mutually recognized certification for open models.
Openness also has commercial edges. In August, EastFrontier reported that Alibaba planned to charge large commercial users of an upcoming open-weight Qwen model, following a licensing approach Moonshot AI had used for Kimi K3. Tsai’s Wuzhen speech did not address licensing terms.
The summit turned the message into an official initiative
The conference released the Action Initiative on Jointly Building Global Open AI (《携手构建全球开放人工智能行动倡议》), which calls for safety and trust to be designed into open AI ecosystems from the start, according to the 21st Century Business Herald. The event’s scale gave the theme a wide audience, with an exhibition of more than 670 companies from 54 countries and regions showing over 1,000 new technologies and products, MyDrivers (快科技) reported. Geely chairman Li Shufu, Sohu chief executive Zhang Chaoyang and Unitree founder Wang Xingxing attended the opening alongside Tsai.
For investors, the measurable follow-through is in Alibaba’s next releases. The derivative-model count is Alibaba’s preferred proof that open distribution builds an ecosystem, and the license attached to the next flagship Qwen model will show how far the company’s "only path" argument extends to its most capable systems.