Just 24 hours after the release of the Qwen3.5 small model series, the core of Alibaba's AI lab has fractured.
Who Left
On March 5, Junyang Lin, the chief AI researcher who led the Qwen model series from zero to over 6 million global downloads, announced his resignation. He was not alone:
- Binyuan Hui — head of Qwen code models
- Bowen Yu — head of post-training
- Kaixin Li — head of Qwen 3.5 and vision-language models
- Several more junior researchers resigned the same day
Lin posted on X that he was "stepping down" without explanation. Chinese tech media 36Kr later reported the trigger: an internal restructuring that would place a researcher recruited from Google's Gemini team in charge of Qwen's technical direction.
A Bizarre Timeline
The day before the departures, the team had just released the Qwen3.5 small model series. After the release, Elon Musk posted on X, describing the models as having "quite high intelligence density," sparking significant community discussion.
Twenty-four hours later, the core team that built those models collectively submitted their resignations.
Leaving at the moment of greatest achievement is unusual in any company.
Alibaba's Response
CEO Eddie Wu accepted Lin's resignation and sent an internal memo:
"Stagnating technically means falling behind. Advancing foundation models is a core strategic priority."
He also announced the formation of a "Foundation Model Special Task Force," co-led by himself, CTO Zeming Wu, and Alibaba DAMO Academy President Jingren Zhou, to coordinate group-wide resources and accelerate foundation model development.
The message was clear: "People leave, but the work continues." Alibaba also reaffirmed its commitment to increasing open-source investment — a critical point, as the architect of Qwen's open-source strategy was the departing Junyang Lin.
Impact on Qwen
Lin's departure is more than a title change.
Qwen's rise from an internal project to a regular on open-source leaderboards in just over a year was driven by a key decision: uniformly adopting the Apache 2.0 license with zero restrictions on commercial use. This strategy drove significant international adoption. It was not a technical decision but a directional one — and the person who made that call has just left.
Will the new leader, with a Google Gemini background, continue this open-source path? Will the technical direction shift? Will the release cadence slow? These are all unknowns.
In practice, Qwen3.6 Plus was still released shortly after the personnel upheaval, so the short-term release pace hasn't broken. Alibaba has invested tens of billions in foundation models — an investment too large to be halted by a few departures.
But losing the open-source lead, code model lead, post-training lead, and vision-language model lead simultaneously means accumulated expertise and methodologies in these areas will take time to rebuild. Whether the upcoming Qwen4 series can maintain its previous pace will only become clear in six months.
Broader Context
Alibaba is not the first Chinese AI lab to experience such a dramatic personnel shakeup, and it likely won't be the last. ByteDance, Baidu, and Tencent have all seen similar turnover of core technical talent.
The typical pattern for Chinese AI giants is: recruit the strongest researchers, iterate rapidly, then adjust leadership during commercialization. In this process, the people who achieve the first breakthroughs are often not the same ones who later take over commercialization — and the seeds of divergence are sown from that point.
The departure of Lin and his team may simply be another iteration of this cycle.
Sources: Alibaba's chief AI developer quits,CocoLoop, taking key team members with him (The Decoder); Did Alibaba just kneecap its powerful Qwen AI team? Key figures depart in wake of latest open source release (VentureBeat); Alibaba's Qwen tech lead steps down after major AI push (TechCrunch)