There are two kinds of AI model. Claude and ChatGPT are restaurants: you order, they cook, but you never enter the kitchen, the recipe, the brain, stays hidden. These are 'closed' models. Meta's Llama and Alibaba's Qwen are chefs who publish the recipe: they release the model's brain (the weights) as a file anyone can download and run on their own server. That's an 'open-weight' model, and download counts are its popularity score. Meta's Llama was the king, until Alibaba's Qwen flooded the market with free models from laptop-size to server-grade, taking the top spot in 2025 and cementing it with a billion downloads by April 2026. The fight is over how. In June 2026 Anthropic filed a complaint with the US Senate alleging that operators tied to Qwen ran a massive 'distillation' attack on Claude, bombarding a strong model with questions and harvesting the answers to copy its skill. Anthropic claims 25,000 fake accounts pulled more than 28.8 million exchanges between April 22 and June 5, subscribing to Claude Code Max on stolen cards to capture even the reasoning traces. Alibaba hit back, banning internal Claude use from July 10 over an alleged tracking 'backdoor.' With the copyright status of model outputs still murky, Anthropic chose politics over a lawsuit.
Why it matters · The question is whether the open-weight crown was won by talent or by a cheap shortcut. Because the target is US-listed Alibaba, the verdict lands not just on Qwen's legitimacy but on sentiment around Alibaba's stock.