Alibaba Appoints Liu Daheng as Head of Qwen LLM Team, Succeeding Lin Junyang

Alibaba has appointed Liu Da Yiheng as head of the Qwen LLM project at the ATH business group's Token Foundry, succeeding former head Lin Junyang. This appointment means that after Lin Junyang announced his departure on social media in March of this year—when Zhou Jingren, then Alibaba Cloud CTO and head of the Tongyi Lab, took over the Qwen model top seat on an interim basis—the Qwen top seat, vacant for about half a year, finally has a clear owner.

Liu Da Yiheng was not an outside hire. After Lin Junyang's departure, in addition to his original responsibility for pretraining, he also took interim charge of the post-training and Coding teams, reporting to Zhou Jingren. In other words, Qwen's three core links—pretraining, post-training, and Coding—were in fact all once in his hands. From interim charge to formally taking full control, this was a path of gradual expansion of authority, not a sudden leap.

It should be noted that the report did not specify the exact date on which Liu Da Yiheng formally took full control of Qwen, nor did it disclose the specific reasons for Lin Junyang's departure or his subsequent whereabouts. These gaps mean that what the outside world sees at present is only the appointment outcome, not the complete decision-making process.

To understand the background of this appointment, it is necessary to turn the clock back to June. Alibaba announced the merger of the Tongyi Large Model business unit with the Future Life Lab to establish the Token Foundry business unit, directly overseen by group CEO Wu Yongming. This adjustment moved the organizational level at which Qwen sits upward, placing it directly within the group CEO's management radius.

At the same time, Zhou Jingren moved to the role of Alibaba Chief Scientist, leading the establishment of the AI Future Research Institute, focused on frontier research. From interim charge of the Qwen top seat to a shift toward frontier research, Zhou Jingren's role change and Liu Da Yiheng's succession constitute two sides of the same organizational adjustment.

However, the report did not provide the organizational size or responsibility details of the Token Foundry business unit, nor did it disclose in detail the actual reporting relationships and division of authority and responsibility of the Qwen team after its establishment. The complete picture of the organizational structure still awaits more information.

Liu Da Yiheng's connection to Qwen can be traced back to the project's earliest days. According to the report, after joining DAMO Academy in 2021, he was responsible for Qwen's earliest pretraining work, subsequently led the development of the Qwen1 through Qwen3.5 series of language models and the Qwen-Math series, was a core participant in the development of models including Qwen-Coder, VL, and Omni, and has cumulatively participated in the open-sourcing of more than 300 Qwen models.

Judging from this résumé, he has in fact been present throughout Qwen's entire development process, from early R&D to continuous iteration. This continuity of being "present throughout" is the practical basis on which he was able to gradually take over the three core links after Lin Junyang's departure.

But it must be flagged that the above résumé data—such as the number of papers and the number of open-source models—all come from statements by the reporting party and lack first-party source corroboration. They can serve as a reference for understanding his technical background, but should not be regarded as independently verified conclusions.

What first brought Liu Da Yiheng into the public eye was a news item in 2020: after multiple rounds of selection, he became the first student in the Southwest region to be selected into the first tier of Huawei's Genius Youth program, with an annual salary as high as 2 million yuan. According to the report, only 5 students nationwide were selected into the first tier that year.

His academic background was likewise highlighted: a native of Zigong, Sichuan, he studied at Sichuan University's School of Computer Science for his bachelor's, master's, and doctoral degrees, under Professor Lü Jiancheng, with research directions in natural language generation and understanding; he has published more than 100 papers in top conferences and journals including ACL, ICML, NeurIPS, EMNLP, and AAAI, with over 60,000 Google Scholar citations, and received the NeurIPS 2025 Best Paper Award as a corresponding author—the other corresponding author being none other than former head Lin Junyang.

After graduation, he joined Huawei's core R&D team, the Cloud Voice and Semantic Innovation LAB, working on NLP technology exploration. Together, these labels form the material through which the outside world understands his "pedigree," but they likewise are statements by the reporting party and have not been independently verified by first-party sources.

Beyond the appointment, the public arrangements at the Alibaba Yunqi Conference released a more intuitive signal. The guest section of the official website placed Liu Da Yiheng in the center position, immediately after Joseph Tsai and Wu Yongming, with the title of Head of the Qwen LLM Project, Token Foundry, Alibaba ATH Business Group.

At the opening ceremony, following Wu Yongming's keynote, he delivered a 15-minute keynote titled "Qwen: Toward Real-World Agents." From the guest ordering to the speaking slot, this arrangement was interpreted by the report as meaning "the degree of emphasis is self-evident."

What needs to be distinguished is that the guest ordering and the speech length are publicly verifiable arrangements, whereas the judgment of "degree of emphasis" belongs to the reporting party's interpretation. The two should not be conflated.

The speech topic "Real-World Agents" was inferred by the report to be the next direction for Qwen under Liu Da Yiheng's leadership. This inference has its reasonableness: the topic itself points directly to agents, and Lin Junyang's first long essay after his departure was also about reflections on the agent era—two successive heads touched on the same direction in public expressions.

But it must be made clear that this is currently only an inference at the level of a speech topic, with no product roadmap to corroborate it. Between a topic and an actual product direction lie multiple links, including model capabilities, engineering implementation, and commercial choices.

At the same time, the Qwen that Liu Da Yiheng has taken over is not in an easy competitive position. The latest version, Qwen3.8 Max, ranks seventh on the Artificial Analysis Intelligence Index, behind only Anthropic's and OpenAI's flagship models and Meta's Muse Spark 1.3. The report did not give specific scores or evaluation details, and the version and evaluation criteria on which the index is based are also unclear. In such a position, the new head still needs to continue delivering model results.