On March 4, 2026, just after midnight Beijing time, Lin Junyang wrote two words on X: “Bye Qwen.” Alibaba’s youngest P10-level executive, a shy, bespectacled 32-year-old who had served as the primary technical architect of the company’s open-source AI model family for years, was confirmed to have filed his resignation the day before. At 1:00 PM on the same day, the CEO of Alibaba was performing damage control in front of an emergency all-hands meeting at Tongyi Lab.
The company’s level of seriousness over the departure was indicated by the quickness of what came next. Alibaba Group CEO Eddie Wu flew in to speak to the group. His message was that Qwen remained Alibaba’s top focus and that the reorganization represented an expansion rather than a retreat. Additionally, he expressed regret for what he called inadequate communication regarding the distribution of compute resources, which is a quite unusual step for a top leader at a company this size. In the same conference, Zhou Jingren, the head of Tongyi Lab, admitted that resources had been limited. In a moment that seemed remarkably open considering the circumstances, he also acknowledged that he had experienced internal marginalization. The chief HR officer flatly refused to answer a question about Lin’s potential return.
After completing his studies in computer science and languages at Peking University, Lin immediately joined Alibaba in 2019. In the ensuing years, he emerged as the main force behind the development of Qwen, a model family that has amassed over 600 million downloads and produced over 170,000 derivative models throughout the world’s open-source community. He had been actively promoting the launch of the Qwen 3.5 small model series just two days before to his resignation. This release garnered industry attention and was apparently liked by Elon Musk. The sudden departure, from someone who had been publicly applauding his team’s efforts just 48 hours prior, struck particularly hard.
There were other people departing besides Lin. On the same day, Yu Bowen, who had overseen post-training research on the Qwen models, left. One of the main designers of the Qwen-Coder series and the head of Qwen Code, Hui Binyuan, had already departed in January to join Meta. Kaixin Li, a PhD candidate from the National University of Singapore and a key contributor to Qwen 3.5, also departed. On March 4, a number of younger researchers sent in their resignation letters. Many people’s emotions were encapsulated in a colleague’s social media post, “Qwen is nothing without its people.”
The internal reorganization that put a researcher hired from Google’s Gemini team above the current Qwen leadership appears to have been the structural trigger, according to reports from a number of Chinese tech outlets, including 36Kr and LatePost. This move went against the team’s preference for the kind of vertically integrated, full-stack approach that Lin had advocated. Alibaba’s overarching strategy goal has been to become more commercially segmented, with various teams owning distinct AI stack layers instead of a single, coherent organization managing it from start to finish. That change seemed challenging for a team that had developed its identity on open-source principles and strong internal control over the model development process.
Alibaba took swift action to close the disparities. After joining Tongyi Lab earlier in 2026, Zhou Hao, a former Senior Staff Researcher at Google DeepMind, took over as head of post-training work, answering to Zhou Jingren. To manage AI development at the group level, Eddie Wu and the company’s two senior CTOs organized a new task force. In February, there were 203 million monthly active users of the Qwen consumer app, up from 31 million in January. The business reported these figures as proof that the model’s momentum had not broken.
That might be accurate. However, it’s probable that the figures only provide a partial picture. The consumer app, not the research pipeline, is reflected in that size of monthly active user growth. It is more difficult to measure and replace what the Qwen team developed over a number of years under Lin, including its open-source positioning, its reputation in the international developer community, and the culture that drew top researchers to Hangzhou rather than San Francisco. It was evident that what Lin stood for went far beyond Alibaba’s organizational structure as the AI community responded to his two-word post in real time, with hundreds of responses from researchers at businesses all around the world expressing gratitude. The dilemma the industry is currently facing is whether Alibaba can maintain that position with a reconfigured team and new leadership.
