Alibaba's Qwen group of AI researchers have been among the many most prolific and well-regarded by worldwide machine studying neighborhood — transport dozens of highly effective generalized and specialised generative fashions beginning final summer time, most of them totally open supply and free.
However now, simply 24 hours after transport the open supply Qwen3.5 small mannequin sequence—a launch that drew public reward from Elon Musk for its "spectacular intelligence density"—the undertaking’s technical architect and a number of other different Qwen group members have exited the corporate beneath unclear circumstances, elevating questions and issues from all over the world concerning the future route of the Qwen group and its give attention to open supply.
The departure of Junyang "Justin" Lin, the technical lead who steered Qwen from a nascent lab undertaking to a worldwide powerhouse with over 600 million downloads, alongside two fellow colleagues — employees analysis scientist Binyuan Hui and intern Kaixin Li — marks a unstable inflection level for Alibaba Cloud and its function as a global open supply AI chief.
These three Qwen Workforce members introduced their departures on X immediately, although they didn’t share the explanations or whether or not or not it they had been voluntary. VentureBeat reached out to sources at Alibaba for extra data and can replace once we receive it. Lin himself signed off with a easy put up: "me stepping down. bye my beloved qwen."
Whereas the corporate celebrates a technical triumph, the sudden exit of its core management suggests a deepening rift between the researchers who constructed the fashions and a company hierarchy now pivoting towards aggressive monetization.
The departing researchers' closing reward: pocket-sized intelligence
The Qwen3.5 small mannequin sequence (starting from 0.8B to 9B parameters) represents a closing masterstroke in "intelligence density" from the founding group.
The fashions make use of a Gated DeltaNet hybrid structure that permits a 9B-parameter mannequin to rival the reasoning capabilities of a lot bigger programs.
By using a 3:1 ratio of linear consideration to full consideration, the fashions preserve a large 262,000-token context window whereas remaining environment friendly sufficient to run natively on normal laptops and smartphones — even in net browsers.
Lin, a PKU humanities graduate and polyglot, has lengthy advocated for this "algorithm-hardware co-design" to bypass compute constraints—a philosophy he detailed on the January 2026 Tsinghua AI Summit.
For the developer neighborhood, Qwen3.5 wasn't simply one other replace; it was a blueprint for the "Agentic Inflection," the place fashions shift from being chatbots to autonomous "all-in-one AI employees" able to navigating UIs and executing advanced code.
The enterprise dilemma
For the 90,000+ enterprises at the moment deploying Qwen through DingTalk or Alibaba Cloud, the management vacuum creates a disaster of confidence.
Many firms migrated to Qwen as a result of it provided a "third manner": the efficiency of a proprietary US mannequin with the transparency of open weights.
Alibaba has just lately consolidated its AI efforts into the "Qwen C-end Enterprise Group," merging its mannequin labs with client {hardware} groups. The purpose is evident: transition Qwen from a analysis undertaking into the working system for a brand new period of AI-integrated glasses and rings.
Nevertheless, the reported appointment of Hao Zhou, a veteran of Google DeepMind’s Gemini group, to guide the Qwen group signifies a shift from "research-first" to "metric-driven" management.
Business analysts, together with these cited by InfoWorld, warn that as Alibaba pushes to fulfill investor calls for for income progress, the "open" in Qwen’s open-weight fashions might turn out to be a secondary precedence — just like what we noticed with Meta after the disappointing launch of its Llama 4 AI mannequin final spring, and subsequent reorganization of its AI division, seeing the hiring of Scale AI co-founder and CEO Alexandr Wang and following departure of preeminent researcher Yann LeCun.
Enterprises counting on the Apache 2.0-licensed Qwen fashions now face the chance that future flagships —such because the rumored Qwen3.5-Max—might be locked behind paid, proprietary APIs to drive Cloud DAU (Every day Energetic Person) metrics.
The takeaway? If you happen to worth Qwen's open supply efforts, obtain and protect the fashions now, whilst you nonetheless can.
The "Gemini-fication" of Qwen?
The inner friction at Alibaba mirrors the tensions seen at OpenAI and Google: the "soul" of the machine is commonly at odds with the "scale" of the enterprise. Xinyu Yang, a researcher at rival Chinese language AI lab DeepSeek, captured this sentiment in a stark put up on X: "Exchange the wonderful chief with a non-core folks from Google Gemini, pushed by DAU metrics. If you happen to choose basis mannequin groups like client apps, don’t be stunned when the innovation curve flattens."
This "Gemini-fication"—the shift towards a extremely regulated, product-centric tradition—threatens the very agility that allowed Qwen to surpass Meta’s Llama in by-product mannequin creation. For the worldwide AI neighborhood, the lack of Junyang Lin is symbolic.
He was the first bridge between China’s deep engineering expertise and the Western open-source ecosystem. With out his advocacy, there are fears that the undertaking will retreat right into a "walled backyard" technique just like its Western rivals.
'Leaving wasn't your selection'
The technical brilliance of the Qwen3.5 launch has been overshadowed by the heartbreak of its creators. On social media, the sentiment among the many group members who constructed the mannequin is certainly one of mourning moderately than celebration:
Chen Cheng, a Qwen contributor, explicitly alluded to a compelled departure, writing in a put up on X: "I'm actually heartbroken. I do know leaving wasn't your selection… I truthfully can't think about Qwen with out you."
Li instructed the exit signaled the top of broader ambitions, similar to a deliberate Singapore-based analysis hub: "Qwen may have had a Singapore base, all because of Junyang. However now that he's gone, there's no motive left to remain right here."
What occurs to Qwen's open supply AI efforts from right here on out?
The recognized details are easy: Qwen has by no means been technically stronger, but its founding core has been dismantled. As Alibaba prepares to face traders for its fiscal Q3 earnings report on March 5, the narrative will probably give attention to "effectivity" and "industrial scale."
For the enterprises at the moment excited concerning the 60% price reductions promised by Qwen3.5, the instant future is shiny.
However for the bigger AI neighborhood, the price of that effectivity stands out as the lack of probably the most vibrant open-source lab within the East.
As Hao Zhou takes the reins, the world is watching to see if Qwen stays a "mannequin for the world" or turns into merely a part in Alibaba’s company backside line.
