In the early hours of March 4, 2026, Lin Junyang, the technical director of Alibaba Tongyi Laboratory, announced his resignation. As a key architect of the team’s infrastructure, he led the development of the M6, Qwen 1.5, Qwen 3 series, and multi-modal models. He also promoted the establishment of a “vertical integration” R&D loop to address the challenges posed by the Scaling Law. His resignation stemmed from Tongyi Laboratory’s plan to split the team into horizontal teams, which reduced his management authority and technical decision-making power. Additionally, Alibaba’s introduction of C-end operational metrics for evaluating research teams weakened their motivation for innovation. Furthermore, the strategic conflict between open-source ecosystem development and closed-source commercialization put Lin Junyang in a difficult position. This talent loss has drawn attention to the technical environment challenges under the pressure of computing power in the industry.
Lin Junyang, the technical leader of the Alibaba Qwen team, announced his resignation in the early hours of March 4, ending his role as the core architect of the project. On the same day, Yu Bowen, the person in charge of post-training tasks, also resigned, and Zhou Hao, a new researcher from DeepMind, took over his duties. During Lin Junyang’s tenure, he led the development of the M6, Qwen 1.5, Qwen 3 series, and multi-modal models. He also promoted the team to establish a “vertical integration” R&D cycle to address the challenges posed by the Scaling Law. His resignation was due to Tongyi Lab’s plan to split the team into horizontal teams, thereby reducing his management authority and technical autonomy. Additionally, Alibaba’s introduction of C-end operational metrics for evaluating research teams weakened their motivation for innovation. Furthermore, the strategic conflict between open-source ecosystem development and closed-source commercialization put Lin Junyang in a difficult position. This talent loss has drawn attention to the technical challenges under the pressure of computing power in the industry.