On the evening of March 3, a social media post sent ripples of surprise through the global AI open-source community.

"Stepping down. Farewell, my beloved Qwen." On the early morning of March 4, Junyang Lin, the technical lead of Alibaba’s Qwen large model team, posted this succinct message—just over a dozen words—on social media.

Dramatically, only 48 hours prior, his team had open-sourced the Qwen3.5 series of compact models, earning personal praise from Elon Musk for their "impressive intelligence density." By coincidence, on the same day, AI Product Rankings released its latest data, showing Qwen surging to third place globally among AI applications, boasting 203 million monthly active users (MAU) and an astonishing 552.83% growth rate.

Departures at the zenith of success often carry a poignant edge.

A Quiet Exit Announced in the Dead of Night

Junyang Lin’s resignation was sudden and understated—no lengthy farewell, no grand vision for the future, just a handful of words.

His colleague and fellow Qwen core contributor, Chen Cheng, commented: "Leaving wasn’t your choice." This remark, like a stone tossed into a still pond, set off a cascade of questions. If departure wasn’t a personal decision, what forces lay behind it?

The timeline leads back to the evening of March 2. The Alibaba Qwen team had just open-sourced four compact Qwen3.5 models, supporting visual understanding and reasoning mode switching, requiring just 7GB of memory for local operation. Elon Musk liked the post on an overseas social media platform, and Junyang Lin reposted it with gratitude. Everything seemed ideal—technological breakthroughs, industry recognition, and high team morale.

Yet, just two days later, this 32-year-old tech prodigy chose to say goodbye.

According to multiple sources, this leadership change is not an isolated incident. Besides Junyang Lin, core members like Binyuan Hui and Kaixin Li have also departed, signaling a major transition in the Qwen team’s leadership.

Junyang Lin’s background is unique in the AI field. Born in 1993, he studied computer science as an undergraduate at Peking University before switching to linguistics and applied linguistics for his master’s degree—a "crossover" that later proved foundational to his research on multimodal large models.

After joining Alibaba DAMO Academy through campus recruitment in 2019, he contributed to projects like the M6 multimodal pre-training model and the OFA unified multimodal framework. In late 2022, when Alibaba integrated the DAMO Academy AI team into Alibaba Cloud to form the Tongyi Laboratory, Junyang Lin, at just 29, was appointed technical lead of Tongyi Qwen.

Over the next two years, he guided the team in fully open-sourcing the Qwen series, spanning models from 0.5B to 72B parameters. The Qwen3-Max, launched in 2025, ranked among the global top three. At the AGI-Next Summit in January, he proposed the concept of "models as products," stressing that foundational model research requires a product manager’s mindset for real-world applications.

Promoted to P10 at 32, Junyang Lin was one of Alibaba’s youngest technical leaders and a recognized "tech star" in the open-source community. Under his leadership, the Qwen series achieved remarkable milestones: over 600 million global downloads, more than 170,000 derivative models, surpassing Meta’s Llama to become the world’s largest open-source model family.

The departure of a tech visionary often marks a turning point.

After news of his resignation broke, discussions in overseas open-source communities were overwhelmingly positive.

AI researcher Nathan Lambert called it a "legendary run." Open-source tool teams like Unsloth AI posted tributes, thanking Qwen for its contributions to the open-source ecosystem. One top community comment cut to the heart: "Qwen is nothing without its people."

Yet concerns were equally real: many developers worried that without its core figures, the Qwen3 series might become "the last masterpiece for a long time," and the team’s strategy could shift from open-source-first to closed commercialization.

Yuchen Jin, CTO of Hyperbolic Labs, also noted that without this core group of innovators, Tongyi Qwen would likely abandon its cutting-edge open-source path and pivot to closed commercial products to meet short-term financial targets.

These concerns are not unfounded. Over the past year and a half, key technical leaders such as former Qwen technical head Chang Zhou, former speech team head Zhijie Yan, and former multimodal and vision lead Liefeng Bo have all left. This latest mass exodus of the core team has thrust the stability of Alibaba’s AI talent system into the spotlight.

The Clash Between Open-Source Ideals and Commercial Realities

Junyang Lin’s resignation brings to the forefront a long-standing industry question: how can open-source large model teams thrive within corporate evaluation systems?

According to industry insiders, Alibaba Cloud recently adjusted its evaluation metrics for the Qwen team—using consumer-app indicators like daily active users (DAU) to measure a foundational model team. Critics argue that this approach of "managing foundational research with app-building logic" may stifle innovation.

The deeper conflict lies in the tension between open-source influence and commercial viability. The Qwen series has seen massive downloads on Hugging Face’s open-source platform, excellent academic citations, and a strong developer reputation, but open-sourcing itself does not directly generate cloud revenue. When team KPIs are tied to commercialization metrics, the dilemma of "greater open-source visibility, harder paid conversion" emerges.

A source close to the team commented: "Qwen has built a strong brand in the tech community, but this may not count as a ‘hard metric’ in internal evaluations."

This misalignment in evaluation methods has been criticized as "treating foundational model research teams like app development teams," directly constraining the team’s frontier innovation space.

The clash between idealists and tech "bureaucrats" has always been a recurring theme in tech companies.

Notably, Junyang Lin’s resignation coincided with Alibaba’s AI brand consolidation. On March 2, Alibaba announced it would unify its AI brands under "Qwen" (Chinese: "Qwen Large Model"), retiring the "Tongyi Qwen" brand.

From renaming the Quark AI glasses to "Qwen AI glasses" to integrating B2B and B2C brands, Alibaba is playing a bigger game. However, frequent personnel changes add uncertainty to this strategy.

Finding a balance between "youthful innovation" and "organizational stability" is a critical challenge for Alibaba’s AI organizational development. At the AGI-Next Summit earlier this year, Junyang Lin said, "If your goal isn’t to help all of humanity, you shouldn’t build large models." This statement may be the best tribute to the spirit of open-source.

Regardless of his personal journey, the Qwen series has already left a significant mark on the global open-source large model landscape. For Alibaba, this is a "gear-shifting moment" that requires careful navigation; for the open-source community, it is a "stress test" worth observing.

Technology evolves, but the belief in technology—held by its people—is the most irreplaceable asset.

In the world of algorithms, parameters can grow infinitely, and models can iterate endlessly, but what truly drives innovation are the souls bold enough to ask "why." Junyang Lin’s seven words are not just the farewell of a tech genius but also a mirror reflecting the core challenge China’s AI industry must confront amid its rapid advance: when commercial metrics become the sole standard of measurement, are we losing those most precious "code poets" who dare to burn for their ideals?