Alibaba's youngest P10 suddenly steps down, AI trend has shifted: top talent is hard to retain, open-source spirit is difficult to monetize

On March 4th in the early morning, Alibaba’s Qianwen large model team leader Lin Junyang announced his departure from the team. This AI talent, who graduated from Peking University with a background in both liberal arts and sciences, previously led the open-source efforts for the Qianwen series of large models and was also Alibaba’s youngest P10-level technical expert.

His sudden departure has sparked widespread industry discussion, reflecting two unavoidable realities in the AI industry: the high frequency of talent turnover and the awkward situation faced by the open-source ecosystem.

Interdisciplinary Background

Just before Lin Junyang stepped down, the Qianwen team announced two major developments: first, the open-sourcing of four small-sized models that can run smoothly on laptops and other mid- to low-end devices; second, Alibaba declared that its B2B large models and C-end application brands would be unified under the name “Qwen” (English: Qwen), replacing the original “Tongyi Qianwen.”

The day before, the team was fully focused on deploying the latest small models, and the next day, Lin suddenly announced his departure. This dramatic turn of events mirrors Lin’s academic journey: a typical science student who graduated with a computer science degree from Peking University, but then “abandoned science for liberal arts,” switching to the School of Foreign Languages to study linguistics and applied linguistics.

Perhaps it was this solid foundation in the humanities that infused his research with more linguistic genes, ultimately guiding his focus toward natural language processing and multimodal learning, allowing him to “get on board” during this wave of AI development. In 2023, Lin became a core technical figure for the Tongyi Qianwen large model, earning the title of Alibaba’s youngest P10 leader and becoming a top industry expert. Additionally, he is an active member of the global developer community and is regarded as a key promoter of the open-source Qianwen large model.

Lin Junyang bids farewell to Alibaba Qianwen.

He has not commented on his future plans or the reasons for his departure.

However, industry sources suggest two possible underlying causes: first, the plan to split the Qianwen team into separate groups for pre-training, post-training, and multimodal tasks, with Lin advocating for closer collaboration across these stages and opposing excessive division; second, the team is about to welcome a technical leader from Google, with significant disagreements on technical directions.

But these speculations have not been confirmed by Alibaba, and the Qianwen team has not responded to inquiries.

Talent War

The intense industry upheaval following Lin Junyang’s departure highlights two unavoidable issues in the AI sector: the high turnover rate of talent and the awkward position of the open-source ecosystem.

This is not the first time Alibaba has lost key personnel. Over the past two years, as the Qianwen large model gained prominence, several technical leaders have left: former chief of large model technology Zhou Chang was recruited by ByteDance; programming lead Hui Bin moved to Meta; Yan Zhijie, head of the Tongyi speech team, left, followed by Bo Liefeng, head of the visual team.

Lin Junyang.

This “talent dilemma” is not unique to Alibaba; globally, the AI industry is embroiled in a fierce “battle for talent.” AI giants are offering sky-high salaries to attract top talent.

For example, Meta spares no expense in recruiting top AI researchers and engineers, with compensation packages often reaching tens of millions of dollars, even up to $100 million. Yet, even with such generous offers, Meta still faces the challenge of talent poaching. A few months ago, renowned Chinese-American AI scientist Pang Ruoming moved from Apple to Meta, reportedly earning over $200 million in total compensation. Surprisingly, Pang did not stay long and recently joined OpenAI.

International competition is equally fierce. Singapore announced this year the launch of a “Top Talent Visa” to attract leading AI experts worldwide.

Domestic tech giants are also eager for AI talent. According to the Ministry of Human Resources and Social Security, China’s AI talent gap exceeds 5 million, with a supply-demand ratio of 1:10. Leading companies like Tencent and Alibaba have seen signing bonuses for core roles in algorithms and foundational model development soar to over one million yuan, and even AI internship daily wages have risen sharply, generally exceeding 500 yuan.

Changing Trends

Although Lin Junyang has not disclosed specific reasons for his departure, some believe his exit is a significant loss for Alibaba’s open-source strategy.

Currently, Alibaba maintains a comprehensive open-source approach, covering large language models, mathematics, programming, speech, and vision, with over 400 models open-sourced, more than 1 billion downloads globally, and over 200,000 derivative models. Nvidia founder Jensen Huang has repeatedly praised Qwen and DeepSeek as the best open-source AI models, and Elon Musk has also lauded Qwen’s small models for their “impressive intelligence density.”

However, amid widespread praise, balancing the high costs of open-sourcing with commercial value has become a difficult challenge Alibaba must face.

Following Lin’s statement, a highly upvoted comment summed up the situation: “If you judge foundational models like consumer applications, it’s no surprise that when the innovation curve flattens, progress slows.”

Latest AI product rankings show that Qwen leads globally with a 552% growth rate, over 200 million monthly active users, making it the third-largest AI application worldwide. Yet, behind the strong promotional efforts, “daily active users” are increasingly becoming the core metric for the Qianwen team, which seems to diverge from the open-source community’s emphasis on decentralization and free innovation.

“Balancing open-source vision and commercial interests is indeed very challenging for companies,” said Jia Yangqing, former vice president of Alibaba’s technology. He noted that conflicts between open-source ideals and commercial priorities are common, citing RethinkDB, an open-source database once beloved by developers but ultimately shut down due to lack of commercial support.

Today, the AI landscape has shifted, and strategic directions need to be reconsidered.

By 2025, large AI models are in the “emergent intelligence period,” where open-source is the mainstream belief among AI talent. Many startups have risen rapidly through open-source ecosystems, with DeepSeek being a notable example. Now, the competition has shifted from frontier R&D to commercial deployment, moving from labs to engineering and monetization. The departure of some technical idealists may be an inevitable choice for large companies during this transition.

“Once the costs of open-source outweigh the benefits, abandoning open-source becomes inevitable,” a developer told reporters. The total cost of ownership (TCO) behind open-source is extremely high, involving not only server and storage expenses but also a large community maintenance team. “For independent developers, open-source is about passion, but for big companies, it’s not just a cost or technical issue—it’s a systemic strategic decision.”

The trend is shifting: since ChatGPT-4, OpenAI has moved from open-source to closed-source development, and even the new open-source “flagship” model from Meta, “Avocado,” is rumored to adopt a closed-source approach.

Open source or closed source? That remains an open question, with no definitive answer yet.