Qwen tops the open-model world with 3 billion downloads
The open-weight center of gravity has a new address: Alibaba's Qwen crossed 3 billion downloads in six months, and Hugging Face's summer ecosystem report crowns it the largest foundation in open AI.
Alibaba's Qwen family has become the most-downloaded open model line on Earth, with more than 3 billion downloads in the past six months — ahead of Meta and Google. Bloomberg reported the milestone, and Hugging Face's "State of Open Models, Summer 2026" report backs it up with ecosystem numbers: 151,448 Qwen-based derivatives — 2.6x Meta's total footprint and 4.7x the number of Llama repos — with the family growing by roughly 200 derivatives a day. HF says Qwen now leads local inference, followed by Gemma, and that while frontier models keep getting bigger, small models still dominate real-world usage: GGUF quantized-model repos are up 464% year over year.
The open-weight center of gravity has moved to China. Two years ago the default fine-tune base was Llama; today HF's report says Qwen "has become part of the default workflow for developers deciding what models to fine-tune and deploy." We covered Qwen's Max-class debut last week — Qwen open-sources Qwen3.8-2.4T-A95B, its first Max-class model — and this download curve is the follow-through. For developers the practical signal is simpler: the biggest open ecosystem now ships from Hangzhou, Apache 2.0 licensed, and free to build on.
AI has become a genuine US midterm issue for the first time, with candidates in roughly 40% of races now posting AI and data-center policies. Washington Post data shows the issue has never before featured so widely in American political life — a shift driven less by frontier-model politics than by local fights over data centers, power, and water. NPR and NBC News report data centers are now a top issue for voters and candidates on both sides, and CBS News covers an Oakland councilmember facing pushback over a proposed downtown data center. Expect AI to keep showing up in attack ads and ballot-box fights, from Virginia to the Bay Area, long before any federal AI law lands.
OpenAI's Computer History — the macOS feature that turns your activity into ChatGPT memories — tracks your clicks and keystrokes, not just app usage. A Verge walkthrough of the feature, which we covered at launch — OpenAI's Computer History turns Mac activity into ChatGPT memories — confirms it logs what you do into a timeline that ChatGPT and Codex can reference. OpenAI says the feature is opt-in, lets you exclude specific apps and sites, auto-ignores incognito tabs, and captures no images, video, or audio — events only. That last point is the key distinction from Windows Recall, which was derailed by screenshots; whether keystroke-level event logs will draw the same scrutiny is the open question, especially since the data doubles as training material for suggested automations.
What to watch: whether Western labs answer Qwen's download dominance with their own open-weight push, and whether Computer History-style tracking draws the same regulatory fire Recall did.
China now runs the biggest open model ecosystem, and OpenAI is logging your keystrokes — is the open-source boom making the privacy tradeoff inevitable? Tell us in the comments.
