Today might be one of the worst days for open-source AI in a long time, and I don’t say that lightly. In this video I break down what just happened to Qwen (Alibaba’s model family) after the core team behind it appeared to get pushed out. That matters because Qwen isn’t just “another” open model line—it's the most downloaded open-source model family in the world, with over a billion downloads and an insane number of variants. More importantly, Qwen has been the team relentlessly shipping small language models (2B–9B-ish) that actually make open weights practical on laptops, phones, and edge devices. I walk through the economics and incentives that make this fragile: companies spend millions training models, then give them away under permissive licenses like Apache 2.0, hoping to win distribution, talent, and ecosystem control. But none of that cleanly shows up as revenue—especially when Alibaba is in a brutal consumer AI war in China and dumping huge money into user acquisition. Then I cover the timeline: a big small-model release, followed immediately by the tech lead and other core contributors stepping down, and reporting that internal reorgs shifted Qwen from a vertically integrated research unit into horizontally managed modules. My takeaway: if Qwen’s open-source output slows or goes proprietary, we get a real gap in small open models (and even speech/ASR) that nobody is positioned to fill right now. If you care about open weights, this is the kind of moment where you download and preserve what matters—because incentives can flip fast.