Qwen 4 release date: Alibaba said on Sep 22, 2026 that Qwen 4 is in training. No launch day, open weights or public API yet. Confirmed vs rumored status.

That is the short version as of October 13, 2026. Below we separate what Alibaba’s own Apsara Conference press release actually said from the rumor mill, explain what a Qwen 4 drop would mean for people choosing open models, and list the signals worth watching.

Qwen 4 release date: confirmed vs rumored

On September 22, 2026, Alibaba Cloud published a press release from its Apsara Conference in Hangzhou. On Qwen 4 it said two things only:

  • Confirmed: the next-generation model, Qwen 4, “is currently in training.”
  • Confirmed as roadmap: upcoming Qwen 4.5 and Qwen 5 series are “projected to scale up to 5 to 10 trillion parameters.”
  • Not announced: any release date, API name, pricing, licence or downloadable weights.

We re-checked today: the Hugging Face Qwen organization still has no repository named Qwen4 or Qwen-4. The latest dense open-weight line people can actually download is the Qwen3.8 family (including Qwen3.8-27B). A search of the official X account @Alibaba_Qwen for Qwen4 / Qwen 4 returned no matching posts. So if a blog gives you a hard Qwen 4 release date, treat it as speculation.

Qwen 4 confirmed vs rumored status table as of October 13, 2026
Status as of Oct 13, 2026. Tier names (Max, Plus, Flash, 27B) appear only in attendee reports, not Alibaba’s written release.

What third parties claim (and why we do not treat it as fact)

Conference write-ups and Reddit threads describe four tiers previewed on stage — Qwen 4 Max, Plus, Flash and a 27B open-weight tier — and some quote the team saying the release would come “very soon.” Those details do not appear in Alibaba’s written press release. Prediction-market odds and leaked YouTube calendars are likewise third-party. We leave them out of the confirmed column, and you should too until a model card, API ID or Hugging Face repo appears under the Qwen org.

What Qwen 4 would mean for people choosing open models

The reason this rumor cycle matters is not the brand name. It is the open-weight habit. Alibaba has repeatedly shipped dense mid-size Qwen models you can run locally (the 27B class across Qwen3.5, 3.6 and 3.8), which is why a rumored Qwen 4 27B gets so much attention in local-LLM communities. If and when a Qwen 4 open-weight model ships, the practical questions are licence, context window, tool-calling quality and whether it beats today’s best open options on your tasks — not a launch-day hype score.

Our take for builders right now: keep your stack model-agnostic. Ship on Qwen3.8, Kimi K2 or whatever wins your evals today, and treat a Qwen 4 drop as a one-line model swap once weights and an API exist. For how other frontier labs are staging gated releases, see our Gemini 4 Argon release date and access guide. For a live open-weight frontier model you can try today, see our Kimi K2 page, and for a wider ranking see best AI model.

What to watch (so you know the day it is real)

  • Alibaba Cloud / qwen.ai blog: a model card or launch post with a date beats every leak.
  • Hugging Face Qwen org: a new Qwen4-* repo is the clearest open-weights signal.
  • @Alibaba_Qwen on X: official account posts, not quote-tweets of rumor threads.
  • API catalog: a Qwen 4 model ID on Alibaba Cloud Model Studio or a major router.

Until one of those appears, “Qwen 4 is coming” is accurate only in the narrow sense Alibaba used: it is in training. It is not released.

Frequently asked questions

When is the Qwen 4 release date?

Alibaba has not announced a Qwen 4 release date. Its September 22, 2026 Apsara Conference press release says only that Qwen 4 is currently in training. As of October 13, 2026 there are no public weights, API name or pricing.

Is Qwen 4 out yet?

No. There is no Qwen4 model on the Hugging Face Qwen organization, and Alibaba’s written statement still describes it as in training. The latest downloadable dense line is Qwen3.8, including Qwen3.8-27B.

Will Qwen 4 be open weight?

Alibaba has not confirmed a licence or open-weight release for Qwen 4. Third-party conference reports mention a 27B tier, but that is not in the official press release. Wait for a Hugging Face model card before assuming you can download it.

What about Qwen 4 Max, Plus, Flash and 27B?

Those tier names appear in attendee reports from Apsara 2026. Alibaba’s official press release does not list them. Treat them as rumored until Alibaba publishes a model card or API catalog entry.

What should I use while waiting for Qwen 4?

Use the current Qwen3.8 models (including Qwen3.8-27B on Hugging Face) or another open model that wins your evals, and keep the model ID configurable so you can switch when Qwen 4 actually ships.