DeepSeek, Qwen and MiniMax Leads Share a Wuzhen Stage to Debate Open-Source Risk

On November 3, Chinese financial outlet Eastmoney reported that the leaders of three of the country's leading large-model companies — DeepSeek, Qwen and MiniMax — appeared together at an "AI Wuzhen Dialogue" session held in the water town of Wuzhen. The headline was blunt about the draw: all three showed up.

A parallel report carried by Sina framed the same session as a joint appearance by the heads of China's top large-model teams, and identified the subject under discussion as how to guard against risk once a model has been open-sourced. Read together, the two headlines make clear the event was not a product launch but a debate about where the safety boundary sits for openly released models.

The three companies occupy different corners of the Chinese AI ecosystem. DeepSeek became widely known for releasing open-weight models, the Qwen family has put open versions into developers' hands, and MiniMax has concentrated on consumer-facing AI products. What they share is the habit of pushing model capability outside their own walls.

Handing out weights means handing out a degree of control. Supporters of open release argue that transparency and reproducibility are preconditions for serious safety research; critics worry that once weights escape, abusers can sidestep any safeguard enforced on a hosted service. Putting that argument in front of three companies that actually publish models is a different exercise from debating it in the abstract.

The timing matters. As the performance gap between open and closed models narrows, openness stops being purely a market strategy and becomes a governance obligation, with release licences, capability evaluations and misuse monitoring all on the table instead of being pushed downstream.

Public coverage so far describes an exchange of positions rather than a set of conclusions, and does not report specific self-regulatory commitments or technical proposals from the three companies. What to watch next is whether gatherings like this one harden into shared open-source safety norms, and whether model cards, usage licences and risk assessments start to converge across China's major labs.