From Conscience to Persona Shock: The Story of Lin Junyang's Resignation from Qwen

In the early hours of March 4th, the global AI community was shaken by a self-written birthday message from Lin Junyang—Chief Technical Officer of Qwen—posted on platform X: “I resign. Farewell, dear Qwen.” This brief message became the climax of a long-simmering internal drama within Alibaba, revealing a far more complex strategic conflict than a mere personnel dispute.

The day before, on March 3rd, Ma Yun and senior Alibaba leaders—including Cai Chongxin, Wu Yongming, Shao Xiaofeng, Jiang Fan—and the Ant Group team gathered at Yun Gu School in Hangzhou to discuss the opportunities and challenges brought by AI—an unmistakable sign of their joint commitment to “All in AI.” Ironically, after this strategic meeting, the decision made in the high-level conference room triggered the departure of one of the key architects of Alibaba’s AI vision.

Strategic Disputes Behind the Self-Resignation

Lin Junyang’s departure was not a simple personal choice but the result of long-standing internal tensions over organizational structure, technological development direction, and business goals. The timeline of events reflects the sharpness of these conflicts.

On the night of March 2nd, Lin led the launch of four small Qwen3.5 models that quickly drew global tech attention. Musk himself praised their “impressive intelligence density,” and Lin responded with optimistic gratitude. On the same day, Alibaba announced the unification of its large B2B and B2C models under the brand “Qianwen,” establishing it as the company’s flagship AI brand.

However, just a few hours after the successful launch on March 3rd afternoon, Lin suddenly left the meeting due to disagreements and submitted his resignation letter. The shock moved some team members of Qwen to tears on the spot. That evening, Lin shared a song titled “Serving a Cup for Myself” on his private WeChat circle—a metaphorical expression of sacrifice and resilience in facing unwanted change.

In the early hours of March 4th, his resignation announcement on X quickly went viral, garnering over 10,000 likes and 1,400 comments in a short time. Industry developers and AI companies like MiniMax flooded in to thank Lin and the Qwen team for their contributions to the global open-source ecosystem.

What was more concerning was what happened next. On the morning of March 4th, Yu Bowen (post-Qwen training leader) and Li Kaixin (key contributor to Qwen3.5/VL/Coder) also announced their resignations. That afternoon, Lin reposted on his WeChat circle: “Sorry, friends, I won’t reply to messages or calls today. I really need to rest. Qwen brothers and sisters, keep going as planned. No problem.”

Chen Cheng, a core contributor to the Qwen team, honestly commented: “I am truly heartbroken. I know leaving isn’t your choice. Yesterday, we still launched the small Qwen3.5 model together. I can’t imagine Qwen without you.” This comment spread industry acknowledgment that Lin’s resignation was not entirely voluntary—an involuntary birthday message to himself, forced rather than a genuine choice.

Alibaba has yet to issue an official public response. However, sources close to Alibaba suggest that Lin may soon upload a statement on X because he was unable to control his emotions at that moment—strong evidence of the involuntary nature of his decision.

Philosophical Tech Conflict: When Ideals Clash with Business Realities

The root of the issue lies in fundamental differences over how to develop large language models. Lin Junyang believes that competitive advantage in large model development comes from deep collaboration across the entire team—from pre-training, post-training, multimodal development, to infrastructure. In contrast, Tongyi Laboratory plans to transform this vertically integrated development model into a horizontal structure, with each module operating independently under direct management.

This structural change would not only narrow Lin’s management authority and business scope but also diametrically oppose his views on the development trend of large language models. Separating workflows would significantly reduce development efficiency and innovation space—a technical argument that also embodies a philosophical struggle.

Beneath this technical dispute lies a deeper battle between open-source approaches and corporate commercial goals. Under Lin’s leadership, Qwen achieved the top spot among global open-source large language models through comprehensive transparency. It became a benchmark for China’s language models penetrating the international market, with impressive data on Hugging Face—the world’s largest open-source AI community—recording over 200,000 derivative models and over 1 billion downloads by January 2026.

However, Alibaba’s priority assessment of Qwen has shifted from technical impact to practical commercial deployment. According to Phoenix Tech reports, some executives viewed the Qwen-3.5 launched on Lunar New Year’s Eve as a “semi-finished product” still requiring refinement. The 3 billion yuan subsidy for Qwen’s consumer applications fell short of expectations, while cloud AI business faced market pressures from aggressive competitors like ByteDance and Tencent.

The misalignment between Lin’s technical ideals and corporate business targets has created a fracture increasingly difficult to mend. In the fiercely competitive AI world, this birthday message to himself symbolizes Alibaba’s identity crisis—whether to be a global tech pioneer or a profit-generating machine.

Alibaba’s Strategic Shift and Challenges Ahead

This personnel change marks a strategic turning point for Alibaba. Since 2025, the company has aggressively recruited top global AI talent. After IEEE Fellow Hsu Chu-Hung joined TONGYI Laboratory, his work overlapped significantly with Qwen’s execution. In early 2026, leading researcher Chou Hao from DeepMind joined, reporting directly to the lab head and taking over post-training responsibilities previously handled by Yu Bowen.

The transition from Lin’s “core-led” model to a “parallel power” structure reflects a broader decentralization strategy. However, successive departures of core Qwen team members have shaken this foundation.

According to internal sources, due to the suddenness of the situation, there are currently no candidates capable of fully replacing Lin. His comprehensive management responsibilities will be distributed across several parallel teams during the reorganization.

As for Lin’s future, industry analysts speculate he may start his own venture or join leading teams in embodied AI or world models. While Alibaba’s superiors are still in contact with him, attempting to persuade him to stay, prospects for reconciliation seem slim.

Over the past three months, the departure of Qwen’s technical lead, post-training head, coding lead, and other core early members has not only affected the development schedule of the next Qwen models but could also trigger a broader talent drain.

Lin Junyang’s resignation and his self-initiated birthday message reflect Alibaba’s comprehensive AI strategy transformation—from a phase focused on technical benchmarks and open-source ecosystem building to a new cycle centered on commercialization. The real challenges ahead include disruptions during development due to loss of core teams, wavering confidence in the global open-source ecosystem, and intensifying industry competition. The ripple effects of this personnel upheaval will test Alibaba’s “All in AI” strategic resilience in the coming years.