Junyang Lin, lead researcher for Alibaba's Qwen AI project, announced his resignation on March 4, 2026, triggering a wave of departures from the core team. The resignations came immediately after the successful launch of Qwen 3.5, one of the most popular open-source model families on Hugging Face, and followed an Alibaba reorganization that placed a researcher from Google's Gemini team in charge of the Qwen project.
Core Team Members Depart Following Reorganization
Lin's resignation message—"me stepping down. bye my beloved qwen"—was followed by departures from multiple key team members who built the Qwen architecture:
- Binyuan Hui: Led code development and the Qwen-Coder series
- Bowen Yu: Directed post-training research for Qwen-Instruct models
- Kaixin Li: Major contributor to Qwen 3.5, VL, and Coder versions
- Multiple young researchers also resigned the same day
The reorganization that triggered the departures installed a researcher from Google's Gemini team as the new Qwen leader, a management change that apparently prompted Lin and others to leave. Alibaba CEO attended an emergency meeting following the resignations, indicating the company recognizes the severity of losing this concentrated talent.
Timing Follows Successful Qwen 3.5 Launch
The resignations occurred immediately after Qwen 3.5's successful release, which features model sizes ranging from 0.8B to 397B parameters with impressive performance across coding, reasoning, and multimodal tasks even at smaller scales. Elon Musk praised the release for its "impressive intelligence density." Qwen has become the most-forked open-source model family on Hugging Face, demonstrating strong developer adoption.
The timing suggests the reorganization was particularly poorly timed, coming just as the team achieved a major breakthrough. Unsloth, a popular fine-tuning tool, published a dedicated Qwen 3.5 fine-tuning guide that reached 263 points on Hacker News, illustrating the model's popularity among developers.
Industry Implications for China's AI Landscape
The departures threaten continued innovation on one of the most successful open-source model families competing with Western companies. If the Qwen team joins competitors or launches a new venture, it could fragment China's AI landscape or significantly strengthen rivals.
Simon Willison's blog post "Something is afoot in the land of Qwen" documenting the crisis reached 502 points on Hacker News with 238 comments, reflecting strong community concern about the future of the Qwen project. The situation echoes the OpenAI board drama but may have longer-lasting consequences since these researchers designed and built the entire Qwen architecture from the ground up.
Key Takeaways
- Junyang Lin and multiple core Qwen team members resigned March 4, 2026, immediately following Qwen 3.5's successful launch
- Departures triggered by Alibaba reorganization placing a Google Gemini researcher in charge of the Qwen project
- Qwen is the most-forked open-source model family on Hugging Face with sizes ranging from 0.8B to 397B parameters
- Alibaba CEO attended emergency meeting, recognizing severity of losing concentrated talent that built entire Qwen architecture
- Team's departure threatens continued innovation on one of the most successful open-source models competing with Western AI companies