(Yicai) March 6 — The recent departure of Lin Junyang, the technical leader of Alibaba Group Holding’s Qwen team, is due to the recent departure of Lin Junyang, the technical leader of Alibaba Group Holding’s Qwen team, a person close to the large-scale language model developer who still faces challenges in the LLM field told Yicai. Junyang’s recent departure is closely related to the company’s internal competition mechanism, which sets up different teams to work on the same high-stakes projects.
Before leaving, Lin was in “competition” within the company with Stephen Hui, a prominent artificial intelligence scholar who joined the Chinese e-commerce giant in February last year, the people said. Zhou Hao, a former senior researcher at Google’s DeepMind who was hired by Alibaba earlier this year, could be the next candidate as he also participates in the company’s internal competition system.
Hui was a tenured professor at the Singapore Management University before joining Hangzhou-based Alibaba in February 2025 as a principal researcher in Alibaba’s Intelligent Information Business Group. His main responsibilities were multimodal foundational models and agent-related research for AI-to-consumer applications. In September 2025, Mr. Hui moved to the company’s Tongyi Lab to focus on multimodal interaction models.
Mr. Hui and Mr. Lin both reported directly to Alibaba Cloud’s chief technology officer, Zhou Jingren, the people said. However, Mr Hui’s internal race results were reportedly less than ideal, while Mr Lin’s Qwen team took delivery of the Qwen 3.5 series during last month’s Lunar New Year holiday.
“Racehorse” is an internal competition mechanism employed by several large Internet companies. Simply put, when a company faces a new project with high uncertainty or strategic importance, different teams within the company can start at the same time and consider different approaches.
The person said Zhou Hao is a strong candidate for the next internal race. Mr. Zhou Hao reports directly to Mr. Zhou Jingren, and the addition of another leader at the same level has intensified internal competition, putting him in a similar situation as Mr. Hui.
On the same day as Mr Lin’s resignation, Yu Bowen, Kwen’s post-training director, also announced his resignation. Additionally, Hui Binyuan, head of Qwen Code, had already joined Meta in January.
In an internal email dated March 5, Eddie Wu said Alibaba “continues to adhere to our open source model strategy and strengthen our efforts to hire talented people.”
Commercialization challenges
Analysts suggested that strengthening talent density is essential as Alibaba has not yet established a strong position in the LLM field. Although Qwen has shown strong performance in open source ecosystem metrics, it faces challenges in terms of commercialization.
Since Qwen was first open sourced in April 2023, it has been downloaded more than 1 billion times worldwide and has more than 200,000 variants, according to data from Hugging Face. This makes it the most downloaded open source model in the world and the model with the most derivatives, far surpassing DeepSeek’s 50 million+ downloads and 2,000 derivatives.
But when it comes to model usage, a more direct market metric, Qwen doesn’t lead. For the week from February 16th to February 22nd, the top five models in OpenRouter’s weekly ranking were MiniMax M2.5, Moonshot Ai’s Kimi K2.5, Zhipu AI’s GLM-5, and DeepSeek V3.2, but Qwen was not included. In the OpenClaw AI agent usage chart, K2.5 consistently maintains the top position, with Qwen 3.5 unable to break into the top ranks. Its large open source ecosystem has not yet translated into real-world usage benefits.
A source close to Alibaba Cloud told Yeatsai that Alibaba’s final view is that the model alone is not enough to form a technological moat, and that ultimately it will be infrastructure. Therefore, the commercial value of the model is likely to come through cloud services focused on selling computing power. Open source can help build the global influence this business model requires.
However, according to IDC data, in the first half of last year, Volcano Engine led China’s AI cloud market with a market share of 49.2%, while Alibaba Cloud’s share was only 27%. Despite Qwen’s vast open source ecosystem, this did not lead to market leadership for Alibaba Cloud.
Editor: Kim Taylor
