Less than 24 hours after the release of the Qwen 3.5 small model, which received praise from Elon Musk for its "amazing intelligence density," the Alibaba Tongyi Qianwen team experienced a major personnel shake-up. Just after the Lantern Festival, the Tongyi Qianwen team suffered a major reshuffle of core personnel: technical lead Lin Junyang resigned, along with three other key figures—Qwen post-training lead Yu Bowen, Qwen Code lead Hui Binyuan, and Qwen 3.5 & VL & Coder core contributor Li Kaixin. This was not just an ordinary departure of a technical lead, but a systemic conflict concerning organizational structure, resource allocation, and open-source strategy. Biteye attempts to reconstruct the full picture of this personnel earthquake and ask a more fundamental question: In the AI era, how should large companies place their technological ideals? I. Overnight Bloodshed: The Collective Departure of Core Members Less than 24 hours after the release of the Qwen 3.5 small model, which received praise from Musk for its "amazing intelligence density," Lin Junyang, the technical lead of Alibaba Tongyi Qianwen, posted a brief farewell message on X in the early morning. As of press time, the post has garnered over 11,000 likes and 4.5 million views, with the comment section filled with heartbreak. Lin Junyang, Alibaba's youngest P10-level technical expert at 32, is gone. Lin Junyang's resume is a typical example of China's new generation of AI technical talent: born in 1993, he earned his undergraduate degree in Computer Science from Peking University, but chose Linguistics for his master's degree; this experience, different from that of AI elites, may have given him an extraordinary intuition for multimodal and semantic understanding. He joined DAMO Academy in 2019, leading the development of OFA and Chinese CLIP; in 2022, he became the head of Tongyi Qianwen; and in 2025, at the age of 32, he became the youngest P10 in Alibaba's history. Three others followed him: Yu Bowen, the head of training for Qwen, also resigned; a few hours later, Hui Binyuan, the head of Qwen Code, posted "me too" and changed his profile to "former Qwen"; a few hours after that, Li Kaixin, a core contributor to Qwen 3.5, VL, and Coder, also announced his resignation and changed his Twitter profile to "Pre Qwen". This star team, which had created over 1 billion downloads globally, over 200,000 derivative models, and consistently ranked first in the open-source large model charts, seems to be disintegrating at a visible speed. II. Reasons and Dilemmas: The Game Between Individuals and Large Companies in the AI Era A tweet from Qwen team member @cherry_cc12 revealed the tip of the iceberg of this turmoil. As information from internal meetings gradually leaked out, we pieced together the full picture of this mass exodus. 2.1 Organizational Dilemma: From Special Forces to Assembly Line The original Qwen Lab should have been a sharp team of tech geeks, each one a special forces soldier and a jack-of-all-trades; Lin Junyang was like a reinforced company commander, leading everyone into battle.However, rumors circulate online that the Qwen team plans to split up, transforming from a "vertically integrated" system encompassing different training processes and modalities into separate horizontally divided teams for pre-training, post-training, text, and multimodal tasks. This is essentially the approach of traditional internet companies—Alibaba may believe that: the early Qwen lab was an internal incubation project, and now, for large-scale application, each stage needs to be broken down into SOPs to improve local efficiency and thus optimize overall efficiency. But this approach is outdated in the AI era. Compared to the high-impact practice of OpenClaw driven by a single person, the rules of the game for AI innovation have changed. 2.2 The Resource Dilemma: Is it there or not? On one hand, "Qwen is the most important thing for the group," while on the other hand, executives say "resources are hard to satisfy everyone." This contradictory statement is similar to management rhetoric of making empty promises. "Qwen is the top priority," "We've done our best as a Chinese CEO," "Resource bottlenecks are due to information transmission process problems"—there are only two possibilities behind this: First, senior management doesn't actually value Qwen, and the investment is only based on AI FOMO; second, there are disagreements among senior management, with the party that doesn't value it continuing to obstruct progress. The result is that even the product lines touted as the highest priority cannot secure basic resources. 2.3 The Game Between Individuals and Platforms: Who Can Rise Above the Organization? The most disheartening leaked information comes from an HR representative: "We can't elevate them to a pedestal; the company can't accept irrational demands to retain them at any cost." Is this true? AI companies are already fiercely competing for talent: In 2024, Zhou Chang, the former technical linchpin of Qwen, quietly joined ByteDance's Seed team after leaving to start his own business, receiving a "sky-high offer" of 4-2 level positions plus an eight-figure annual salary; in 2025, Meta offered a $200 million compensation package to snatch Pang Ruoming from Apple, including high stock options and milestone incentives linked to technological breakthroughs. Did the HR conduct competitor research? Is this statement wrong? It also subtly aligns with traditional Chinese organizational philosophy: individuals cannot rise above the organization. But the reality is that in the AI era, the bargaining power of super-individuals is unprecedented, even redefined by the platform. 2.4 Political Struggle: Whose Party Are You On? Internally, it was claimed that "political factors were not considered throughout the process," yet it was mentioned that "it was necessary to consider where to place Zhouhao for maximum efficiency." The subtext is intriguing: the key is not whether things can be accomplished, but whether one is "obedient"; in large companies, a superior's sense of security sometimes carries more weight than a subordinate's actual ability. 2.5 The deeper tension stems from the misalignment between open source and commercial paths. Qwen has established a huge reputation in the global open source community—its download volume, number of derivative models, and international recognition are all high. However, open source does not bring direct users and revenue.As Qwen grew, the group naturally asked, "I've invested so much, shouldn't you give me some return?" III. Reflection: The AI Dilemma of Large Companies This happened at Alibaba, which isn't surprising. The documentary "The Annual Meeting Can't Stop" is modeled after Alibaba, and one classic line perfectly encapsulates this: "If you can't solve the problem, solve the person who raised it." Alibaba's logic seems to be: Qwen will keep running even without anyone. "What we're doing is very ambitious; 100 people are definitely not enough, we need to expand"—this statement not only shows that Alibaba doesn't understand AI, but also that AI doesn't understand Alibaba. Web3 next door was also amused. In the internet era, platforms empowered individuals, pursuing standardization, process-orientation, and replicability; individuals depended on platforms, and platforms defined the rules. The AI era is evolving into: super individuals possess stronger bargaining power, even defining platforms in reverse. AI innovation relies on small teams, high density, and rapid iteration—a "special forces model." When large companies use the organizational logic of the internet era to manage the creativity of the AI era, conflict is almost inevitable. Behind the organizational chaos lies a collective confusion among large companies about how to manage geniuses. When HR asks employees, "What price do you think you're willing to pay?", those who truly have the potential to shape the future have already voted with their feet. [Biteye]