Report: Alibaba teases Qwen4 — Max, Flash, Plus, 27B due next week
A slide and a secondhand 'next week' are all that's official — but tooling is already prepping and a Qwen4-architecture preview is downloadable now.
Copy markdown
The lineup, straight off a slide
At Apsara Amsterdam on Oct 29, Qwen's Xudong Guo showed a "Qwen4 Series Coming Soon" slide naming Qwen4-Max, Qwen4-Flash, Qwen4-Plus and Qwen4-27B. No weights, model card, API ID or pricing exist yet — the only evidence is attendee photos.
"Next week" is a quote, not a release date
Attendees report Alibaba Cloud's CTO said Qwen4 ships "next week." It's a secondhand stage remark, not a press release or blog post — so plan for it, but don't wire it into anything until weights or an API ID actually land.
The tooling is already prepping
The hardest signal isn't the slide — it's the code. qwen-code PR 15762, merged Oct 28, sizes the "qwen4 generation" at a 1M-token context in the CLI, and inference runtimes are reportedly staging support too. Groundwork like this usually lands days before weights.
What you can run today
If you want to kick the tires before launch, Qwen3.8-Flash-Next is open-weight on Hugging Face and billed by the Qwen team as "an early preview of the architecture used in Qwen4." It's the only Qwen4-flavored thing you can actually download right now.
Don't buy the 10-trillion headline
Ignore "Qwen4 = 10T params" write-ups: Alibaba's 5–10T figure is for the later Qwen4.5 and Qwen5, per its September Apsara Hangzhou release. A 27B is in the Qwen4 lineup, but whether it ships with open weights is still unconfirmed.