On March 5 Alibaba Group CEO Wu Yongming sent an internal note formally approving the resignation of Lin Junyang, a high‑profile technical lead of the Qwen (千问) open‑model project, and announced a foundation‑model support group to be led by himself, Alibaba Cloud CTO Zhou Jingren and Fan Yu. The memo reaffirmed Alibaba’s commitment to an open‑model strategy and pledged continued investment in AI research, while placing Tongyi Lab under the stewardship of existing cloud leadership.
The personnel shift followed a tense three‑day sequence. On March 2 Lin published small‑footprint variants of Qwen 3.5 on X; on March 3 he reportedly clashed with management at an internal meeting and submitted his resignation; and on March 4 he posted a brief farewell on X and told his team he needed to rest. The episode culminated in an emergency all‑hands at Tongyi Lab and the unusually public confirmation of a senior tech departure by Alibaba’s chief executive.
Interviews with three people who know Lin or use Qwen paint a picture of a classic technical founder‑figure leaving not because of scandal but because of an organizational mismatch. Colleagues describe Lin as a focused, highly technical engineer who rose through Alibaba via campus recruiting and direct laboratory work rather than corporate politicking. To many in China’s developer community, he has been the face and steward of Qwen’s open‑source ethos — someone who answered technical questions, wrote detailed documentation and pushed frequent model releases.
That personal connection explains the strong reaction among users and startups that built products on Qwen’s freely published weights. Longtime enterprise users say Qwen’s rapid iteration and the decision to make high‑quality weights available set it apart from closed rivals; those users now fear a slowing of open releases or a shift in priorities as corporate and product interests reassert themselves. The concern is not merely sentimental: open‑model weights represent a material subsidy of compute and energy that large firms must choose to give away.
The episode also highlights a familiar tension playing out across the AI industry: the clash between engineering ideals and product or business imperatives. CEOs and founders interviewed for this story said the AI era intensifies that tension because model development demands longer horizons, bigger capital and different metrics of success. Engineers prize leaderboard performance, reproducibility and community trust; companies must balance those against go‑to‑market speed, customer demands and the economics of maintaining costly infrastructure.
Alibaba’s swift internal response — elevating senior management to lead a support group and publicly reaffirming open‑model commitments — will calm some investors and enterprise customers. Yet the optics are poor for an organisation that has used open‑source releases to build a developer ecosystem; the departure of a visible technical leader risks eroding the informal ties that sustained that ecosystem even if formal policy remains unchanged.
Finally, Lin’s exit underscores how scarce senior model‑engineering talent remains globally. Public invitations from external teams and signals from peers suggest Lin will not lack options, and that other firms and venture investors are watching closely. For Alibaba, retaining the technical culture that produced Qwen as an open, fast‑moving project may be a harder task than preserving the banner of open‑sourcing itself.