The landscape of open-weight AI models has been jolted by a series of high-profile departures from Alibaba’s Qwen team. Just as the group celebrated the release of the Qwen 3.5 family—a suite of models praised for their efficiency and power—technical lead Junyang Lin announced his resignation. This departure has triggered a domino effect, with several key researchers responsible for code development, post-training, and multimodal capabilities also stepping down from the Tongyi Lab.

The timing is particularly striking given the technical success of the Qwen 3.5 series. The family includes a massive 397B parameter model alongside highly optimized smaller siblings that punch well above their weight class. For instance, the 2B parameter variant remarkably includes both vision and complex logic capabilities in a footprint small enough to run on consumer hardware. These models have quickly gained traction in the developer community for their strong performance in coding tasks, often rivaling much larger proprietary systems.

Speculation suggests the exodus stems from a recent internal reorganization, where a former researcher from a competing lab was reportedly placed in a leadership position above the existing team. Alibaba Group CEO Wu Yongming has already intervened, holding an emergency meeting to stabilize the laboratory and potentially retain remaining talent. While the future of the Qwen roadmap remains uncertain, the impact of these researchers—who achieved world-class results with fewer resources than Western competitors—will likely shape the next phase of global AI development wherever they land.

The landscape of open-weight AI models has been jolted by a series of high-profile departures from Alibaba’s Qwen team. Just as the group celebrated the release of the Qwen 3.5 family—a suite of models praised for their efficiency and power—technical lead Junyang Lin announced his resignation. This departure has triggered a domino effect, with several key researchers responsible for code development, post-training, and multimodal capabilities also stepping down from the Tongyi Lab.

The timing is particularly striking given the technical success of the Qwen 3.5 series. The family includes a massive 397B parameter model alongside highly optimized smaller siblings that punch well above their weight class. For instance, the 2B parameter variant remarkably includes both vision and complex logic capabilities in a footprint small enough to run on consumer hardware. These models have quickly gained traction in the developer community for their strong performance in coding tasks, often rivaling much larger proprietary systems.

Speculation suggests the exodus stems from a recent internal reorganization, where a former researcher from a competing lab was reportedly placed in a leadership position above the existing team. Alibaba Group CEO Wu Yongming has already intervened, holding an emergency meeting to stabilize the laboratory and potentially retain remaining talent. While the future of the Qwen roadmap remains uncertain, the impact of these researchers—who achieved world-class results with fewer resources than Western competitors—will likely shape the next phase of global AI development wherever they land.