Qwen Team Exodus: Alibaba Loses 3 Key AI Leaders (2026)

Three senior leaders have left Alibaba's Qwen team in Q1 2026 — including tech lead Lin Junyang, the architect behind the world's most downloaded open-source AI project. The departures came days after Qwen released 9 models in 16 days, right as the platform crossed 1 billion downloads on HuggingFace (Domain-b, March 2026). Qwen's leaders chose different paths — what it means for the open-source AI community

> TL;DR: Qwen tech lead Lin Junyang, post-training head Yu Bowen, and staff researcher Binyuan Hui all left Alibaba in Q1 2026. The departures happened during Qwen's most productive stretch ever — 9 models in 16 days, 1B+ downloads, and 203 million monthly active users. A colleague hinted the exit was involuntary. All already-released models remain available and fully functional.

Who Left the Qwen Team?

Three key figures departed Alibaba's Qwen AI division within the first quarter of 2026. Each played a different but critical role in making Qwen the most downloaded open-source model family in the world.

Lin was the public face of Qwen. He joined Alibaba in 2019, became Qwen tech lead in 2023, and steered the project from a lab experiment to 1 billion+ downloads globally (OfficeChai, March 2026). With 42,000+ Google Scholar citations, he was one of the most-cited AI researchers in China.

Yu Bowen ran post-training — the process that turns raw model weights into useful instruction-following models. Binyuan Hui focused on code generation, one of Qwen's strongest selling points for developers.

Additional contributor Kaixin Li also departed, stating: "Grateful for the chance to work with such brilliant minds. Proud of our impact" (Benzinga, March 2026).

What Triggered the Departures?

The resignations appear to be involuntary. Qwen contributor Chen Chang wrote on X: "I know leaving wasn't your choice. Just last night, we were side by side launching the Qwen3.5 small model" (OfficeChai, March 2026). Kaixin Li was more direct: "there's no reason left to stay."

TechNode reports that the exits coincide with "internal organizational adjustments" at Tongyi Lab, including plans to reorganize model development teams (TechNode, March 2026). Bloomberg notes Lin had publicly warned about the gap between Chinese AI labs and OpenAI — a position that may not have aligned with Alibaba's corporate messaging (Bloomberg, March 2026).

Wenting Zhao, a research scientist still on the Qwen team, called it "the end of an era" (TechCrunch, March 2026).

The Ironic Timeline

The departures happened at the absolute peak of Qwen's momentum. Here is the full sequence of events.

That is 9 new models in 16 days, followed by the lead architect walking away. The Qwen mobile app surged from 31 million monthly active users in January to 203 million in February — a 550% increase (Domain-b, March 2026). Qwen became the third-largest AI product globally, behind only ChatGPT and Doubao.

VentureBeat asked the question on everyone's mind: "Did Alibaba just kneecap its powerful Qwen AI team?" (VentureBeat, March 2026).

What Does This Mean for Open-Source AI?

Every model already released stays released. Open-weight model distributions are permanent — Alibaba cannot retract the weights, licensing, or model cards for any published Qwen model. The 400+ models on HuggingFace, the 170,000+ derivative models, and the 1 billion+ downloads remain intact (OfficeChai, March 2026).

The risk is about what comes next. Lin Junyang set the open-source philosophy that made Qwen unique among Chinese AI labs. Without him, Alibaba could take several directions:

• Stay the course — The Qwen brand is too valuable to shut down. Alibaba recently unified all AI branding under "Qwen," dropping the old "Tongyi Qianwen" name.
• Slow open-source releases — New leadership might prioritize commercial API revenue over community releases.
• Shift to partial openness — Release smaller models openly but gate flagship models behind APIs.

For context, Alibaba has committed over $53 billion to cloud and AI infrastructure. The Qwen project is central to its cloud strategy. Shutting it down entirely would be corporate self-harm.

Should You Still Run Qwen Models Locally?

Yes. The models on your machine — or available through Ollama — work exactly as well today as they did last week. Here is what has not changed:

• Qwen 3.5 Small (0.8B-9B) still runs on 2-14 GB RAM
• Qwen 3.5 Medium (27B-122B) still fits on 16-72 GB machines
• Qwen 3.5 Flash still delivers million-token context
• All models still carry permissive open-source licenses

If you are running Qwen locally with Ollama, nothing changes. The weights are on your machine. No API call, no server dependency, no risk of discontinuation.

# These still work perfectly
ollama run qwen3.5:4b
ollama run qwen3.5:9b
ollama run qwen3.5:27b

The risk is only forward-looking: will future Qwen models maintain the same quality and openness? That remains uncertain until Alibaba names a successor and signals its direction.

What Happens to Qwen Next?

Alibaba has not named a successor to Lin Junyang. The Qwen team still has researchers and engineers — the departures are concentrated at the leadership level. Three scenarios are plausible.

Scenario 1: Internal promotion. Alibaba promotes from within the existing Qwen team. Institutional knowledge is preserved, but the creative direction may shift. This is the most likely outcome given Alibaba's track record. Scenario 2: External hire. Alibaba recruits a new tech lead from another Chinese AI lab or from overseas. This could take months and carries the risk of cultural mismatch with the existing team. Scenario 3: Absorption into Alibaba Cloud. The Qwen team loses its semi-independent status and gets folded into Alibaba Cloud's broader engineering org under CTO Zhou Jingren. Open-source output would likely slow under this structure.

Regardless of which path unfolds, the AI talent market in China is extremely competitive. DeepSeek, Baidu, ByteDance, and Moonshot AI all want exactly the kind of researchers who just left Alibaba. Lin Junyang's next move will be closely watched by the entire industry.

Qwen by the Numbers (Before the Exodus)

FAQ

Will existing Qwen models stop working?

No. All released Qwen models are open-weight. The files exist on HuggingFace, Ollama, and your local machine. There is no server callback, no license revocation, and no kill switch. They work indefinitely.

Why did Lin Junyang leave?

No official reason has been given. A colleague's post suggests it was not voluntary — "I know leaving wasn't your choice." TechNode reports internal team reorganizations at Tongyi Lab. Bloomberg notes Lin publicly criticized the gap with OpenAI, which may have created friction with management.

Is Qwen still the best open-source model for local AI?

As of March 2026, yes. The Qwen 3.5 family — especially the 4B and 9B — scores at or above much larger models on MMLU-Redux, LiveCodeBench, and agent benchmarks. No other open-source family matches Qwen's breadth across model sizes.

Should I switch to a different model family?

Not yet. The Qwen 3.5 models you have today are excellent. If Alibaba's next release cycle shows quality regression or license changes, that would be the time to evaluate alternatives like Llama, Gemma, or Mistral. For now, check which Qwen model fits your Mac.

Where might the departed leaders go next?

No announcements yet. China's AI talent market is intensely competitive. DeepSeek, Moonshot AI, ByteDance, and well-funded startups are all actively recruiting senior AI researchers. A new venture is also possible — Lin has the citations, reputation, and track record to attract funding easily.