Even before the festive vibes of the Spring Festival had completely faded away, China's AI large model industry witnessed a major personnel shake-up that sent ripples throughout the sector.

In the early hours of March 4, 2026, Junyang Lin, the technical lead of Alibaba's Tongyi Qianwen (Qwen), posted a succinct English message on overseas social media: "I'm stepping down. Farewell, my beloved Qwen.", officially announcing his resignation.

Following suit, Bowen Yu, the leader of the post-training phase for Qwen, and Kaixin Li, a key member of the core team, both confirmed their departures. Adding to the turmoil, BinYuan Hui, the head of Qwen Code, who had already left to join Meta in January 2026, the core technical layer of Tongyi Qianwen underwent rare and significant upheaval.

More notably, on the same day as Lin's potential resignation (March 3), Alibaba founder Jack Ma convened with senior executives, including Joe Tsai and Eddie Wu, at Yungu School to deliberate on the challenges and opportunities presented by the AI era.

Fundamental Disagreements Surface

On one hand, the founder's resolute stance on AI's future, and on the other, the departure of core technical team members, have drawn considerable attention to Alibaba's large model strategy.

As one of Alibaba's youngest P10 technical experts, Junyang Lin was the driving force behind the development of Tongyi Qianwen from the ground up.

He spearheaded the entire process of Qwen series model R&D, iteration, and open-source strategy, propelling Tongyi Qianwen to become a mainstream open-source large model with over 600 million downloads globally and top-tier community influence.

Bowen Yu oversaw the post-training phase, a crucial stage for aligning model capabilities and optimizing performance, directly influencing user experience and product competitiveness. Kaixin Li was instrumental in model iteration.

The simultaneous departure of three core leaders was no mere coincidence. The Tongyi Qianwen team had faced similar situations before: in July 2024, the inaugural technical leader Chang Zhou left to join ByteDance; visual algorithm expert Liefeng Bo and speech recognition expert Zhijie Yan followed suit.

These individuals were all deeply entrenched technical pillars at Alibaba, well-versed in the underlying architecture, R&D rhythm, and core secrets. Their continuous attrition dealt a significant blow to team stability, technical continuity, and R&D momentum.

The primary catalyst for this talent exodus was fundamental disagreements over organizational restructuring and technical routes.

Currently, the Qwen team operates under Tongyi Lab, led by Alibaba Cloud CTO Jingren Zhou, adopting a "vertically integrated" structure where pre-training, post-training, multimodality, and infrastructure are uniformly coordinated under Lin's oversight.

This highly interconnected, rapid-response model was instrumental in Qwen's swift ascent to the top tier of global open-source large models.

Diverging Paths

However, Tongyi Lab recently charted a new course, planning to split the integrated team into multiple horizontally divided teams—pre-training, post-training, text, multimodality—each functioning independently and reporting directly upward.

This adjustment not only significantly curtailed Lin's managerial authority but also contradicted his steadfast technical philosophy.

Lin had consistently emphasized that large model R&D necessitates deep end-to-end collaboration, cautioning that fragmentation would drastically reduce R&D efficiency, prolong iteration cycles, and escalate communication costs.

As early as mid-2025, cracks had begun to appear. The Qwen team attempted to establish its own infrastructure, seeking to break free from Alibaba Cloud PAI's unified support to safeguard R&D autonomy and efficiency—a move that clashed with the group's resource coordination strategy.

This structural split became the final straw. The simultaneous departures of Yu, Li, and others were interpreted by the industry as a collective stance against the technical route and organizational changes.

When a tech company dismantles a top-tier model team with assembly-line divisions and constrains cutting-edge R&D with management KPIs, the exodus of talented individuals becomes almost inevitable.

As if misfortune came in threes, amid the talent collapse, Tongyi Qianwen also found itself in a precarious position in market competition.

During the 2026 Spring Festival, the AI industry witnessed its most intense red-packet war, with ByteDance's Doubao, Alibaba's Qianwen, Tencent's Yuanbao, and Baidu's ERNIE Bot investing billions in subsidies to capture user mindshare.

Leveraging ecosystem scenes like Alipay, Taobao, Hema, and Fliggy, Tongyi Qianwen launched the "Qianwen Treats" campaign, using freebies, subsidies, and scene-linked incentives to aggressively acquire users. Daily active users (DAUs) surged from around 17 million pre-campaign to a peak of 73.5 million, a more than fourfold short-term increase.

But after the initial frenzy, the data revealed a stark reality: DAUs plummeted post-subsidy, with user retention, engagement depth, and brand recognition falling short of expectations. The gaps in model capabilities, multimodal experience, and ecosystem coverage with industry leaders remained unaddressed.

This costly traffic acquisition strategy failed to translate into sustainable competitive advantages.

Besieged on All Fronts

Under multiple pressures, Tongyi Qianwen found itself under siege on all fronts.

At the talent level, rebuilding the core team is an arduous task. On the R&D front, iteration rhythms may slow down, and collaboration efficiency faces challenges. In the market, user stickiness is weak, open-source homogenization is severe, and commercial monetization is fragile. Coupled with stricter global regulations and rising compliance costs, profitability prospects are even more uncertain.

Despite Alibaba's deployment of an "AI golden triangle" combining Tongyi Lab, Alibaba Cloud, and T-Head, emphasizing its status as one of only two companies globally (alongside Google) with top-tier capabilities across large models, cloud, and chips, the technological foundation cannot mask the real-world dilemmas of frontline team turmoil and execution fractures.

Jack Ma's remark that "AI brings immense impact, but also vast opportunities" serves as both a sober assessment of industry trends and an urgent reminder for Alibaba itself.

With AI technology evolving rapidly on a weekly basis and industry competition intensifying, large models are swiftly transitioning from technical benchmarking to a decisive phase of ecosystem and commercialization.

For Tongyi Qianwen, stabilizing the core team, aligning technical routes, and efficiently delivering technical and commercial value are now urgent priorities.

Only by maintaining organizational stability can it continue refining model capabilities and safeguard technological advantages. Only with a consistent strategic direction can it foster synergies in open-source ecosystems, scene implementation, and computing power support.

And only by swiftly translating technology into user value and market results can it gain a foothold in the fiercely competitive landscape.

This statement serves as both a warning and a call to action, urging Alibaba AI to seize the window of opportunity, respond to industry upheavals with decisive actions, and transform technological confidence into long-term competitiveness.