Qwen jumped from 1% to 69% of open-model fine-tunes

Qwen's share of open-model fine-tunes went from 1% in January 2024 to 69% in February 2026, according to Sequoia's "America's Open-Model Paradox," published July 24. Sequoia's read: most American AI startups now have Chinese open weights somewhere in their stack, and Thinking Machines built its Inkling model on Moonshot Kimi K2.5 synthetic data as one example.

Sequoia's explanation for why is the part that matters. US labs have no legal path to distill their own frontier models, so startups post-train on whatever weights are actually available to them. Right now that's mostly Chinese. Sequoia wants "controlled teacher access" from American labs as the fix.

That number landed in the middle of the loudest open-weights policy week of the year. A 25-company letter, hosted on Nvidia's servers and dated July 24, warns Washington against "premature restrictions" on open-weight models. Signatories include Nvidia, Microsoft, Meta, Mistral, Hugging Face, and Perplexity. OpenAI, Anthropic, and Google didn't sign. Sam Altman said he was "glad to see it" anyway, without adding his name, a straddle that got 566 upvotes worth of dissection on r/LocalLLaMA. Roughly 200 startup founders separately petitioned the Trump administration not to cut off access to Chinese open-weight models, per Politico, pulling over a thousand points on Hacker News.

Andrew Ng, in The Batch, points to a sharper case: when commercial LLMs refused on safety grounds to analyze a set of attack logs, Hugging Face ran the open GLM 5.2 model instead. His point is that a tool's safety comes from how it's used, not how it was built. That example carries extra weight given the breach itself was caused by a closed model.

Worth saying plainly: everyone writing about this has a stake. Sequoia backs startups that benefit from cheap bases. Nvidia sells hardware that runs open weights locally. The companies that didn't sign the letter sell closed models. The 69% figure is the one number in this pile you can actually go check. Go check what's in your own stack before the policy fight gets decided for you.