Lin Junyang leaves, Wu Yongming responds urgently, Alibaba Qwen personnel shake-up within 48 hours

Jiemian News Reporter | Cheng Lu

Jiemian News Editor | Wenshu Qi

For two full days, the AI community at home and abroad has been flooded with news about Lin Junyang’s departure.

The incident happened suddenly. Less than 48 hours after the open-source release of the Qwen 3.5 small models (0.8B/2B/4B/9B), which received Elon Musk’s praise, on the night of March 4th, Lin Junyang announced his resignation on social platform X (formerly Twitter): “me stepping down. bye my beloved qwen.” The post quickly garnered 13,000 likes and nearly 6 million views, sparking global AI community attention and discussion.

Prior to this, Lin Junyang had already been associated with a series of labels—Alibaba’s youngest P10 technical executive, head of the Tongyi Qianwen series large models, the lead and “spokesperson” for Qwen open source.

After Lin Junyang’s departure, multiple core team members followed suit, including Yu Bowen, head of post-training for Qwen; Kaixin Li, a key contributor to Qwen 3.5/VL/Coder; and Binyuan Hui, head of Qwen Code. The future development of Qwen has become a topic of intense speculation.

On the afternoon of March 4th, Alibaba’s Tongyi Laboratory held an emergency all-hands meeting. Alibaba Group Chairman and CEO Daniel Zhang, Chief Talent Officer Jiang Fang, and Alibaba Cloud CTO Zhou Jingren responded to team and organizational adjustments. Senior Alibaba executives emphasized that the foundational model of Qianwen remains the most important project for Alibaba Group. They aim to bring in talent to expand the team, but new recruits will inevitably lead to organizational changes, which may not have been communicated well.

As the news continued to ferment, this morning, Jiemian News learned that Alibaba CEO Daniel Zhang responded directly to Lin Junyang’s departure in an internal email to Tongyi Laboratory.

The email confirmed approval of Lin Junyang’s resignation and conveyed three key signals: maintaining the open-source model strategy; continuing to increase investment in AI R&D and talent recruitment; and establishing a “Foundation Model Support Group” composed of Daniel Zhang, Zhou Jingren, and Fan Yu to coordinate group resources and support foundation model development.

With top-level clarification, Lin Junyang’s departure was settled, but the waves and discussions have not stopped.

Behind Lin Junyang’s Departure

Regarding the reasons for Lin Junyang’s departure, the puzzle pieced together from various responses and Jiemian News’s understanding is roughly as follows. First, the KPI adjustments for the Qwen team this year did occur, but not as the X rumors suggested—using DAU of consumer apps like the Qianwen app to measure the foundational model team.

Jiemian News learned that, in considering the commercialization loop, the group’s frontier research has faced restrictions. Lin Junyang also expressed and discussed this at the AGI-Next summit at Tsinghua University in January.

He pointed out that the core difference in AI industry development between China and the US lies in resources. US companies have 10 to 100 times more computing power, allowing bold investments in next-generation foundational research and exploration of unknown technological fields. Domestic companies have comparatively limited resources.

“Just delivering already accounts for most of our computer capacity,” Lin Junyang said. This has, paradoxically, driven the domestic industry to hone the core ability of deep algorithm and infrastructure synergy optimization.

This deep collaboration mode that Lin Junyang is proud of is also changing. According to 36Kr, previously Qwen had its own pre-training, post-training, and infrastructure teams, with models spanning language, multimodal, and code directions. But in a new round of adjustments, Tongyi Laboratory plans to split Qwen teams into pre-training, post-training, visual understanding, and image teams, merging them with other groups like Tongyi Wanshang and Tongyi Bailin. Without sufficient communication, conflicts erupted.

However, Jiemian News also learned that internally, some voices believe that conflicts at this stage are not necessarily bad. As AI development becomes more competitive, advancing requires higher talent density. Organizational pain is inevitable, and timely adjustments are necessary.

An internal Alibaba source revealed that Lin Junyang’s departure was driven by the group’s strategic shift from the foundational model to overall corporate strategy, prompting the need to recruit more top technical talent, especially those with extensive experience in large-scale model R&D and engineering implementation.

During this process, Lin Junyang’s responsibilities were adjusted. After multiple discussions, he did not accept the changes and voluntarily resigned.

It’s worth noting that, besides organizational issues, Alibaba recently unified its AI branding. On March 2nd, internally, Alibaba consolidated its AI branding under “Qianwen,” covering both the foundational large model (Qwen) and specialized domain models. The Qianwen app is Alibaba’s flagship AI application for consumers. This move aims to eliminate confusion caused by multiple names like Qianwen, Tongyi Qianwen, and Qwen.

The adjustment had already been brewing since the end of last year, reflecting Alibaba’s strategic shift in AI. Recently, Alibaba’s founder Jack Ma appeared twice discussing AI. On February 4th, Ma visited the Qianwen Spring Festival project team at Alibaba’s Hangzhou headquarters. Soon after, Qianwen launched the “Spring Festival Guest List” event, offering free meals and entertainment during the holiday, integrated with Alibaba’s ecosystem. During the festival, user “one-sentence orders” exceeded 200 million, with each token burning representing a significant computational effort.

Strategic shifts, organizational realignment, and resource allocation are questions large tech companies must answer in the AI era.

Talent Wars in the AI Era

Looking back, Lin Junyang’s personnel change was just the trigger, but its ripple effects have sparked discussions about open-source strategies and AI talent competition.

First, industry concerns revolve around the contradiction between open-source ideals and commercialization pursuits within big companies. Compared to ByteDance’s Doubao’s high presence in the domestic consumer market, Qwen’s strength lies in its technical reputation and developer ecosystem, especially its massive downloads on Hugging Face, making it popular among academia and developers.

But open source also means ongoing free community contributions and immense pressure. There are worries about whether Qwen will continue to deliver cutting-edge open-source models or shift toward closed commercial routes. In response, Daniel Zhang’s latest statement aims to dispel such speculation—Alibaba will continue to support open source.

Second, on the AI talent front. Recently, a widely circulated phrase on X is: “Qwen is nothing without its people.” AI scholar Nathan Lambert called Lin Junyang’s departure the end of a “legendary journey”; Unsloth AI sincerely thanked Qwen for its contributions to the open-source community.

However, an insider close to Alibaba told Jiemian News that internally, Lin Junyang is not the core figure of Qwen or Tongyi Laboratory. Although he has long been one of Alibaba’s most active accounts in overseas tech communities, frequently engaging with global developers on GitHub and X, making him the “spokesperson” for Qwen, the true soul of Qwen remains Alibaba partner and CTO of Alibaba Cloud, Zhou Jingren.

Lin Junyang is not the first to leave as a technical lead of Tongyi Laboratory. In 2024, Zhou Chang, who reported to Zhou Jingren and was the former technical lead of the Tongyi Qianwen large model, left to join ByteDance’s seed team, causing labor arbitration disputes and making waves in the industry. After Zhou Chang’s departure, Lin Junyang led the release of Qwen 3 and 3.5 open-source models.

In the future, talent turbulence may become a common challenge for AI giants. During OpenAI’s transition from a research lab to commercialization, similar upheavals occurred. In 2025 alone, over 20 senior executives and top researchers left, including key contributors to GPT-4 and the O1 series models.

As valuation soared and commercialization pressures increased, OpenAI shifted most computational resources and algorithm support toward supporting ChatGPT’s large language models, marginalizing frontier projects like Sora (video generation) and DALL-E (image generation). This mature corporate mechanism diluted the original research spirit and was a core reason for personnel instability. By the end of 2025, only CEO Sam Altman and President Greg Brockman remained from the original 11 founding members.

An investor told Jiemian News that after Lin Junyang’s departure, Alibaba should pay attention to whether a wave of mid-level staff will follow.

Regarding the new team member taking over the post-training work for Qwen, Hao Zhou from Google Gemini has also been widely discussed. With experience at Meta and Google DeepMind, Zhou participated in Gemini 3.0, but his background is more in reinforcement learning. The open-source community remains cautious about his ability to lead Qwen effectively.

Returning to Lin Junyang, the 32-year-old Alibaba’s youngest P10, where will he go next?

Born in 1993, Lin Junyang earned a Master’s in Linguistics from Peking University and joined Alibaba in 2019. In 2020, Alibaba launched the M6 multimodal fusion model project, and with his background in linguistics, Lin was included as a core developer. In 2022, the M6 technical route was established as Alibaba’s core direction for general large models. He was promoted steadily, leading the development of the unified multimodal pretraining model OFA and the Chinese pretraining model Chinese CLIP, eventually becoming the technical lead for the Tongyi Qianwen series. Six years of rapid promotion reflect Alibaba’s heavy reliance on AI talent.

Currently, there is no confirmed news about his next move.

Past trajectories of Tongyi Laboratory alumni include moves to ByteDance, Tencent, JD.com, and other teams. Industry speculation suggests Lin might establish his own new lab, join open-source friendly teams like DeepSeek, or even see Elon Musk’s xAI and Hugging Face “racing to recruit.” Many possibilities exist.

Yesterday afternoon, Lin Junyang posted on social media: “Sorry everyone, I won’t reply to messages or calls today. I really need a rest. Brothers at Qwen, continue as planned, no problem.” Undoubtedly, this highly acclaimed 90s generation figure will continue to attract attention regardless of his next steps.

The storm may gradually subside, but the questions remain. How should large companies balance open-source ideals and commercialization goals? How will the influence and developer reputation built by Qwen’s open-source models be evaluated within corporate systems? In the future, Alibaba will need to reassess its position and strategy in the AI era and face the changing landscape of AI talent competition.

(Contributed by Wang Qiang, Jiemian News)