Alibaba's next model generation, Qwen4, may arrive in early November 2026. On October 29, 2026, at Alibaba Cloud's first Apsara Conference in Europe, held at the Beurs van Berlage in Amsterdam, Alibaba Cloud CTO Feifei Li said "Qwen4 next week," according to a post from attendee John T Davies typed live from the room. A later photo he posted shows a stage slide reading "Qwen4 Series Coming Soon" during a talk by Qwen research scientist Xudong Guo.
Alibaba has not put that date in writing. Its official press release from the event is about a new cloud region in the Netherlands and does not mention Qwen4. So the timeline rests on one spoken line and one slide, both relayed by people in the audience.
What was said and shown
- The quote: Davies attributes "Qwen4 next week" to Li, who is Alibaba Cloud's CTO and president of its international business.
- The slide: it says "coming soon," not a date. Davies' photo shows four names on it, which Nvdium and Pasquale Pillitteri also list: Qwen4-Max, Qwen4-Flash, Qwen4-Plus and Qwen4-27B.
- Not announced: a release date in writing, prices, a license, parameter counts, benchmarks or which models get downloadable weights.
Read literally, "next week" from an October 29 talk means November 2 to 8, 2026, as Nvdium and Pillitteri also note. A slide that says "coming soon" is a softer promise than a date on stage, and Pillitteri notes that a line relayed from the room can slip by a few days. Pillitteri notes that the Qwen team has often released models with an X post and weights already on Hugging Face.
What Alibaba had said before
On September 22, 2026, at the main Apsara Conference in Hangzhou, Alibaba said it was training Qwen4, Reuters reported. It also said the later Qwen 4.5 and Qwen 5 series are expected to scale to 5 trillion to 10 trillion parameters. Its current flagship, Qwen3.8-Max, has 2.4 trillion parameters, according to Reuters. That gave a direction but no date. As Rui Ma noted on X, before Amsterdam all Alibaba had officially confirmed was that the model was in training.
For background on the current lineup, see our Qwen3.8 explainer.
A clue in Alibaba's own code
One technical sign points the same way. In Alibaba's open source terminal coding agent, Qwen Code, pull request 15762, titled "fix(core): size the qwen4 generation at 1M," was merged on October 28, 2026 (UTC). It makes the tool treat every qwen4 model name as having a 1,000,000 token context window, instead of the 262,144 token default those names got before.
The pull request lists the names it tested, including qwen4-max, qwen4-plus, qwen4.5-max, qwen4-coder-plus and qwen4-omni-120b-think. It says the last of those is already used in the project's test files. Two limits apply here. The change is a setting in a coding tool, not a model spec sheet. And its tests ran against a fake local server, not a live Qwen4 endpoint. Its author also writes that the 1M figure was reported by a reviewer, not something the code or the models.dev catalog can confirm. The change shows the Qwen Code project is preparing for qwen4 names. It does not confirm the real window size.
Why the 27B model gets the most attention
Rui Ma wrote that the model people "seem most excited about" is the 27B, "which some think could run on a single gaming GPU." Nvdium expects Max, Plus and Flash to live in the cloud. Alibaba published open weights for Qwen3.8-27B in August. "If it ships open weights, frontier-ish ai runs on your own gpu," wrote @sensei_prek. Nvdium notes that a dense 27B model at 4-bit is a little over 15GB of weights, which fits on a 24GB card such as an RTX 3090 with little room left for long context.
That is an estimate, not a spec. Alibaba has not said whether Qwen4-27B is dense or a mixture of experts model, which changes the memory math, and it has not said any Qwen4 model will be open.
Some posts went further than the sources. One post said "Alibaba just confirmed Qwen4 is landing next week." Rui Ma was more careful: she wrote that the CTO "apparently" said it "according to people in the room," and that "the slide still just says 'coming soon.'"
What Alibaba did announce in Amsterdam
The official news was about infrastructure in Europe. Alibaba Cloud opened its first public cloud region in the Netherlands, in Amsterdam, with two data centres. It is the company's fourth European region after Frankfurt, London and Paris. The company says the expansion brings it to 109 availability zones across 32 regions, and that it is "planning to establish its first cloud regions in Finland, while expanding its data center footprints in Germany and France over the next 12 months."
The release also names European companies working with Qwen models. JetBrains is exploring Qwen models for its Junie coding assistant, "across both cloud and local scenarios," and Amsterdam-based orq.ai is adding Qwen models served from Model Studio, Alibaba's AI platform, with EU data localisation. Separately, the same day, Alibaba Cloud said on X that Qwen3.8-Flash is now available in the European Union on Model Studio.
Why it matters
Qwen4 may not be the only Chinese release that week. Moonshot said on October 23 that full weights for its Kimi K3.1 Pro will be out by November 4, Hello China Tech reported, so two Chinese releases could land in the same week. The Amsterdam event also shows Alibaba selling Qwen to European companies with in-region data storage, for customers who, in the release's words, "require their data to be processed and stored locally in Netherlands and within Europe."
As of October 29, 2026, Alibaba had not confirmed a date in a press release or on its official X accounts, so treat the November 2 to 8 window as reported, not confirmed. The places to watch are the Qwen organization on Hugging Face, the @Alibaba_Qwen account on X, and the official Qwen blog.
Sources
- John T Davies on X: "Qwen4 next week" (Oct 29, 2026)
- John T Davies on X: "Qwen4 Series Coming Soon" slide (Oct 29, 2026)
- Alibaba Cloud: first public cloud region in the Netherlands (Oct 29, 2026)
- QwenLM/qwen-code pull request 15762
- Reuters: Alibaba deepens AI push with new chip, bigger model (Sept 22, 2026)
- Alibaba Cloud: Alibaba Unveils Roadmap on Full-Stack AI Strategy (Sept 22, 2026)
- Nvdium: Qwen4 lands next week in Max, Plus, Flash, 27B
- Pasquale Pillitteri: Qwen4 release next week?
- Hello China Tech weekly recap (Oct 25, 2026)
- Rui Ma on X (Oct 29, 2026)
- Alibaba Cloud on X: Qwen3.8-Flash in the EU (Oct 29, 2026)
- ryo (@notlikeryo_) on X (Oct 29, 2026)
