In the early hours of March 3, 2026, Justin Lin posted six words on X.
https://x.com/JustinLin610/status/2028865835373359513
"I resigned. Goodbye, my dear Qianwen."
Just 24 hours earlier, he was still interacting with Elon Musk on X. Musk praised the Qwen 3.5 small model for its "staggering intelligence density," and he happily replied, "thx elon." The tweet was filled with cheers from global developers.
https://x.com/JustinLin610/status/2028550818035843144
24 hours later, his colleague Chen Cheng posted a tweet:
"I know leaving wasn't your choice."
Post unavailable
Of course, the Qwen code is still on GitHub, and the team still has hundreds of people working; the model won't stop iterating just because a few people left. I know that.
But the vitality of an open-source project has never been just about the code.
Alibaba has issued no official response. But the timeline is noteworthy. As early as December 2025, Alibaba began integrating Qwen into the C-end (consumer) side, establishing the "Qwen C-End Business Group," bringing the Qwen App, Quark, and AI hardware under one organization. On March 2, the day before Justin Lin left, Alibaba announced the unification of "Tongyi Qianwen" under the name "Qwen," clarifying that the Qwen App is the C-end flagship.
https://x.com/fxtrader/status/1998307525062787310
https://x.com/fxtrader/status/1988855599359029507
Commercial transformation. Organizational adjustment. Efficiency first.
I understand these terms.
But then I saw this post:
https://x.com/kuaidaoqingyi/status/2028992502624387562
He said he didn't know the specific details but could fully imagine them.
I don't know the inside story either. Perhaps the truth is simpler; perhaps the strategic shift required a different type of person. But the problem isn't why the person was changed. The problem is what changing this person signifies.
So I want to ask three questions. To Qwen. To the person who made this decision.
First Question: What kind of person have you lost?
Justin Lin, born in '93, Peking University CS undergrad, Master's in Linguistics, became Alibaba's youngest P10 at 32. From the first day Qwen-7B was open-sourced on Hugging Face to over 700 million downloads and 170,000 derivative models, he was the one tweeting every day, replying to issues, and engaging with global developers.
To put it bluntly: the person who made this decision might not truly know what Justin Lin meant to Qwen. I'm not even sure if this was a "decision." Maybe he left on his own, maybe he was reassigned, maybe both. Chen Cheng's "leaving wasn't your choice" is the closest thing to an answer in public information, but it's just a colleague's feeling, not an official statement.
Regardless of the reason, the result is the same: he left. And the things he did don't generate KPIs.
Replying to technical questions from global developers on X every day doesn't generate KPIs. Explaining the design logic of inference frameworks on Hugging Face doesn't generate KPIs. Giving a keynote at ICLR to explain Qwen 2.5's technical details to researchers worldwide doesn't generate KPIs.
But what these things generate is worth ten thousand times more than KPIs.
It's called trust.
Behind 700 million downloads is trust. Behind 170,000 derivative models is trust. Behind Singapore's National AI Program choosing Qwen over Meta's Llama is also trust.
Among all Chinese large model teams, Justin Lin is likely the only one who could consistently interact with the global developer community on English social media. If you raised an issue on Hugging Face, the one replying might be him. If you complained about slow inference on X, the one debating you was also him.
You have lost the most recognizable face of Qwen to the world. That voice. That most active, warm connection with global developers.
I also study fine-tuning and know that the resilience of an open-source ecosystem doesn't depend on one person. A stable release rhythm, clear licensing, a predictable roadmap, and active issue maintenance—these institutionalized elements are the long-term foundation. Qwen 3.5 was released as scheduled around his departure, the Apache 2.0 license remains, and the GitHub repository is still updating. These are facts.
But institutions are the skeleton; people are the flesh and blood. The developer community chooses to trust a project often because they know a specific person.
You might think you can just swap in someone else to tweet. Someone more professional, with better "brand tone," and prettier PPTs.
But the developer community never cares about brands. They care about people.
Look at what Linus Torvalds means to Linux. Look at how many people left when Google shifted from a research culture to the Gemini product line, later spawning a batch of top AI startups. Google's models are still iterating, but 'Google AI's' credibility in the research community has never returned.
Building the trust of a global developer community takes three years. Shaking it can take just one sentence.
But that's just about one person.
Why is this?
Answer me.
Second Question: What kind of path are you tampering with?
I previously wrote an article saying that China's open-source models have become the "Digital Belt and Road" of the AI era.
Literally.
Independent development teams in Eastern Europe, Central Asia, and Southeast Asia now default to Qwen's MoE system for learning inference, fine-tuning, and building products. They lack the deep penetration of US tech giants and have no stable local models. Qwen is cheap, effective, and open-source, making it the most realistic choice. Singapore's National AI Program abandoned Meta for Qwen last year. The ten ASEAN countries are rushing to board the train of China's open-source AI. In the developer ecosystems of 155 Belt and Road partner countries, Qwen's shadow is everywhere.
This isn't about selling APIs. This is purely paving the road.
Once a model becomes someone's default choice, it's no longer just a tool. The architecture determines how compute is allocated, the inference framework determines how products are built, and the MoE sparse activation logic determines how engineers understand "efficiency." You don't need to sign contracts or negotiate partnerships; the technical track itself brings people onto your path.
And this path, objectively, has exceeded the boundaries of a single company like Alibaba.
I know Alibaba is a listed company, not a state-owned enterprise, and has no obligation to pay for national strategy. But the fact is, Qwen's open-source ecosystem has deeply meshed with the national AI layout, whether Alibaba actively sought it or not.
Qiushi.com explicitly stated: We must help ASEAN countries establish sovereign large models and bridge the global intelligence gap.
China goes open-source, the US goes closed-source—this pattern is taking shape. Alibaba has no obligation to serve national strategy. But Qwen's open-source ecosystem has, in fact, become the most important puzzle piece in this pattern.
Every organizational decision you make has a larger spillover effect than you realize.
And Qwen is the vehicle that has traveled the furthest and dug the deepest on this road.
Justin Lin was the driver of that vehicle. Not just writing code. He alone sustained more than just Qwen's overseas community. In my opinion, it's no exaggeration to say he sustained the personified presence of Chinese open-source models globally.
Now you tell me the organization has been adjusted.
I'll say it again. You aren't just moving personnel in a business segment. You are tampering with the most critical node on this path. It connects the trust of global developers in Chinese open-source AI. And this trust happens to be one of the foundations for the AI cooperation China has spent three years promoting under the Belt and Road framework.
Imagine those startup teams in Southeast Asia using Qwen to build products. Seeing the technical lead and core members leave on the same day, they won't think "this is normal organizational optimization." They will do something very quiet: the next time they choose a model, they'll take another look at Llama.
Of course, trust built over three years won't collapse overnight. But it will start to loosen. Not because the model got worse. But because open-source ecosystem users need a basic expectation: the core team of this project is stable, the roadmap is predictable, and the product I build on your architecture today won't lose support tomorrow due to internal restructuring.
When the technical lead and core members leave on the same day without any public explanation or succession plan from the company, a massive question mark is placed over that expectation.
A question mark isn't a collapse. But a question mark needs an answer. An answer provided in a public, institutionalized way. If you don't answer, the question mark will grow on its own.
Yes, everyone knows the logic of an open-source ecosystem is measured in decades. A listed company's decision cycle is measured in quarters. The tension between these two timescales is a contradiction every commercial company building open-source infrastructure must face.
I'm not saying you must sacrifice company interests for a larger ecosystem. You are a listed company and must be responsible to shareholders; I understand that perfectly.
But you should at least know what you are touching. You are adjusting a team. But what this team built is something many people are using and relying on.
And the matter of one person plus the matter of one path points to a larger question.
Why is this?
Answer me.
Third Question: Do you actually still want to win this war?
Who are Qwen's competitors today?
It's not Doubao. Not Kimi. Not Wenxin.
I believe it's GPT. It's Claude. It's Gemini.
Justin Lin said a hard truth at the AGI-Next summit at Tsinghua this January: US compute infrastructure might exceed China's by one to two orders of magnitude. Most of Alibaba's compute is consumed by daily delivery, leaving very limited resources for frontier research.
One to two orders of magnitude.
Under these conditions, how were these achievements made? They were built on algorithmic innovation, architectural design, and trust earned tweet by tweet in the global developer community. To put it more bluntly, they were carved out bit by bit from GPT and Llama.
Now you are taking these achievements to make C-end products. Integrating into Taobao, Alipay, Amap, and Fliggy. AI service functions, ordering takeout with one sentence—nearly 200 million "one-sentence orders" during the Spring Festival.
The numbers look great.
But I want to ask: in these pretty numbers, have you accounted for one thing?
The credit value Justin Lin built in the global developer community.
This number isn't in your KPI system. It's not in the financial reports. It's not in any PPT, right?
But it is the true capital Qwen has to fight GPT.
What do you think the competition with US models relies on? C-end user numbers? Conversion rates for one-sentence orders? The number of Alibaba ecosystem services integrated?
Any fool knows it's not that.
It relies on engineers around the world choosing to stand on your side.
When an Indian entrepreneur decides which model to use as a foundation, he chooses you. When a Brazilian research team decides whose architecture to use for fine-tuning, they choose you.
They choose you not because your App can order takeout in one sentence. It's because a person named Justin Lin seriously replies to every one of their questions on X every day. It's because this person makes them believe that Chinese open-source models are trustworthy.
So why now?
Answer me.
You lost a person. That was Qwen's face to the world.
You tampered with a path. That was the channel for Chinese open-source AI to go global.
Do you still want to win this war against US models? If you do, you must prove through action that this sharp sword hasn't been dulled by a change in personnel.
Finally, I want to say something. Not to Alibaba, but to everyone who still cares about this.
Regardless of how Alibaba adjusts its organization, regardless of how Qwen charges, and regardless of where Justin Lin goes, I will continue to support Qwen.
Not because Alibaba did such a great job.
But because this war must be won.
Chinese AI has been very clear about what it needs to do from day one: on this open-source path, become more useful, more credible, and more irreplaceable than US closed-source models.
Not copying. Not chasing. Proving that open-source can win.
Under conditions of one to two orders of magnitude less compute. Under conditions of chip blockades. DeepSeek used extreme cost-effectiveness to make Silicon Valley rethink efficiency, and Qwen's ecosystem coverage made Llama feel real competitive pressure.
Once this route is set, there is no retreating.
So no matter how Qwen struggles internally, no matter how executives change, no matter what the financial reports say, as long as you are still fighting GPT, fighting Claude, and fighting Gemini, I stand with you.
But I hope the decision-makers think clearly about one thing.
You are holding a sword.
The sword for Chinese open-source AI to face off against the US closed-source system. Forged by the likes of Justin Lin over three years.
This sword is ready to be unsheathed.
GPT-5.3 is iterating. Claude Opus 4.6 is evolving. The Gemini 3.1 team is more stable than ever. No one on the US side is sabotaging their own ranks. They are preparing for war.
And you, on the eve of unsheathing the sword, have replaced the sword-smith.
You can change people. You can adjust the organization. You can do C-end. You can charge fees. Fine.
But this sword must not break.
Because it isn't just your company's sword.
Every developer on this open-source path, every startup team using Qwen to build products, every engineer who chose to stand with China—they are all counting on this sword to win.
So I will only say one last thing.
The sword forged for three years—it's time to unsheathe it.
