Descriptions:
On March 4, 2026, Alibaba’s Qwen project lost most of its founding leadership in a single day. Lin Janzang — technical director of Qwen, born in 1993, and one of Alibaba’s youngest-ever top-ranked engineers — posted his resignation on X at 1:00 a.m. Beijing time with no transition plan. Within hours, the heads of Qwen Code, Qwen post-training, and QwenVL followed, alongside senior researchers Bing Yuan, Bowen Yu, and Kaisen Lee — the people who built the Qwen instruct series, Qwen Coder, and Qwen 3.5VL respectively.
Fahd Mirza, citing sources inside Alibaba, reports that the triggering issue came to light at an emergency all-hands called by CEO Wu Yongming: the Qwen research team was receiving less GPU compute than external Alibaba Cloud paying customers. The Cloud CTO acknowledged the problem at the meeting but offered no concrete remediation, and the departures accelerated from there. Alibaba’s official framing — that this is an expansion and team merger, not a contraction — is undercut by the context: ByteDance’s Doubao is approaching 200 million daily active users while Alibaba’s consumer AI app sits at roughly one-tenth that figure, and the company’s priorities are visibly shifting toward product metrics and monetization.
Lin Janzang built Qwen into the dominant open-source model family globally, releasing over 400 models with a team of just over 100 engineers — compared to ByteDance’s ~2,000. His replacement reportedly comes from Google DeepMind and reports to the Cloud CTO, a structural signal that Mirza argues points toward deprioritized open releases and smaller models, with significant implications for the developer community that has built extensively on Qwen.
📺 Source: Fahd Mirza · Published March 04, 2026
🏷️ Format: News Analysis
