# Qwen 4 Release Date: What Alibaba Has Confirmed So Far > Alibaba said on September 22 that Qwen 4 is in training. As of November 1 there is no model ID, weights or price yet. What is confirmed, and what to use now. - Canonical HTML: https://omniakey.com/blog/qwen-4-release-date - Markdown: https://omniakey.com/blog/qwen-4-release-date.md - Published: 2026-11-01 (updated 2026-11-01) - Author: Kiki, COO at OmniaKey - Language: en - Category: Guide - Reading time: 7 min - All posts: https://omniakey.com/blog/llms.txt --- Qwen 4 is not out yet. As of November 1, 2026 (00:21 UTC), no Qwen 4 model ID, weights or price appears on Hugging Face, Alibaba Cloud Model Studio or OpenRouter. The clearest official statement is [Alibaba's September 22, 2026 press release](https://www.alibabacloud.com/press-room/alibaba-unveils-roadmap-on-full-stack-ai-strategy): Qwen 4 is "currently in training". What you can call today is the Qwen3.8 generation, including [Qwen3.8-Max](https://omniakey.com/qwen3-8-max) and [Qwen3.8-Flash](https://omniakey.com/qwen3-8-flash). **Short answer** - Qwen 4 remains in training, and none of the catalogs we checked on November 1, 2026 lists a Qwen 4 model. - "Qwen 4 Max", "Plus", "Flash", "27B" and the "next week" launch date come from third-party sites. Alibaba has confirmed none of them. - Use Qwen3.8-Max or Qwen3.8-Flash now; both are current Model Studio models with published prices. ## Can you use Qwen 4 today? No. Four of Alibaba's own pages and OpenRouter, all checked on November 1, 2026, showed the same result: - [Hugging Face's Qwen organization page](https://huggingface.co/Qwen) lists no model with "qwen4" or "qwen-4" among its 50 most recently updated; the newest is Qwen-Image-2.1-Turbo, created October 9, 2026. - [Model Studio's models page](https://www.alibabacloud.com/help/en/model-studio/models), last updated October 28, 2026, has nothing starting with `qwen4`; its recommended text models are qwen3.8-max, qwen3.7-plus and qwen3.8-flash, and the China page shows the same three. - [Model Studio's release list](https://www.alibabacloud.com/help/en/model-studio/newly-released-models), last updated October 30, 2026, ends with qwen-audio-3.1-realtime-flash on October 27, 2026. - [OpenRouter's model list](https://openrouter.ai/api/v1/models) ends its Qwen listings with qwen3.8-max-prime, added September 23, 2026. - [GitHub's QwenLM organization](https://github.com/QwenLM) has no Qwen 4 repository. Qwen's own blog and ModelScope build their lists in the browser, so our checks could not read them. There is no Qwen 4 model ID to call yet; the models you can call are in the last section. ## What has Alibaba confirmed? Here is what Alibaba has put in writing about Qwen 4, in its own words. |Item|What Alibaba said|Source|Date| |-|-|-|-| |Status|Qwen 4 "is currently in training"|Apsara Conference press release|Sep 22, 2026| |Roadmap|Qwen 4.5 and Qwen 5 are "projected to scale up to 5 to 10 trillion parameters"|Apsara Conference press release|Sep 22, 2026| |Amsterdam event|No Qwen 4 mention; the event page says "Qwen foundation models"|Press release and event page|Oct 29, 2026| |Preview model|"experimental preview of the architecture that will underpin Qwen4"|Hugging Face model card|Aug 24, 2026| [Alibaba Cloud's blog](https://www.alibabacloud.com/blog/alibaba-clouds-2026-apsara-conference-full-stack-ai-roadmap-along-with-global-market-expansion-plan_603598) repeated the training line and the 5-to-10-trillion roadmap on September 25, 2026. The closest thing to Qwen 4 you can download today is Qwen3.8-Flash-Next. [Its model card](https://huggingface.co/Qwen/Qwen3.8-Flash-Next), created August 24, 2026, describes 125B parameters with 6B active, plus a 51B n-gram embedding and 4B for multi-token prediction, a 262,144-token context that extends to 1,000,000, and the qwen-community-1.0 licence. Its config names the model type `qwen4_exp`. ## What Alibaba has not said yet Six things you would need before planning around Qwen 4 are still missing: - A release date, or even a month. - The Qwen 4 model IDs and parameter sizes. - A price for any Qwen 4 model. - The context window and the output limit. - Whether the weights will be open, and under which licence. - When Qwen 4 reaches the Model Studio API. Until Alibaba answers these, plan with the Qwen3.8 models on the price list. ## Where do the "Max, Plus, Flash, 27B" and "next week" claims come from? Alibaba has confirmed neither the four tier names nor a launch in the week after October 29. Both come from third-party sites, and the dates below are when those sites published them. The site aitoolsreview\.co.uk wrote on September 29, 2026 that Qwen 4 Max, Flash, Plus and 27B were previewed in Hangzhou on September 22; the press release names none of them. pasqualepillitteri.it published the same four names with "zero specs" on September 22, and two GitHub links it gives as proof returned "not found" on November 1, 2026. yottalabs.ai repeated the four names on September 29, 2026 without linking an official page. cellcog.ai, updated October 29, 2026, says an Amsterdam slide showed the four models as coming soon, and that X posts by @jtdavies quoted Alibaba Cloud's CTO saying Qwen 4 arrives next week. flex.ai wrote on October 29, 2026: "Alibaba Cloud's CTO, Feifei Li, said the Qwen4 family will launch the week after the October 29 conference, according to HuggingNews." Alibaba's own [Amsterdam press release](https://www.alibabacloud.com/press-room/first-public-cloud-region-in-the-netherlands) and [event page](https://www.alibabacloud.com/en/apsara-conference-amsterdam) never mention Qwen 4. The week after October 29 is November 2–6, 2026. At 00:21 UTC on November 1, Hugging Face, Model Studio and OpenRouter listed nothing under Qwen 4, so that window stays open and unproven. Qwen's own code uses the name. A Qwen team member merged a [qwen-code pull request](https://github.com/QwenLM/qwen-code/pull/15762) titled "fix(core): size the qwen4 generation at 1M" on October 28, 2026. Inference engines carry `qwen4_exp` code for the preview: transformers has 69 matching issues and pull requests, including [#50138](https://github.com/huggingface/transformers/pull/50138) merged October 30, vLLM 455, SGLang 197, and llama.cpp merged [#30965](https://github.com/ggml-org/llama.cpp/pull/30965) on October 23. None of this code names a released Qwen 4 model. A Hugging Face account called "stealth-model" republishes quantized Qwen3.8-Flash-Next files tagged "qwen4", including one created October 31, 2026. It belongs to a third party, and those files are the preview, not Qwen 4. The "next week" date has also reached developers: on Hacker News on October 31, 2026, @wds [wrote](https://news.ycombinator.com/item?id=50321450): "If you are bound to consumer-accessible hardware, Qwen3.8-27b today and Qwen4-27b when it releases next week". ## If you need a Qwen model this week Pick by job. |If you need|Do this now| |-|-| |The top Qwen3.8 tier|qwen3.8-max, 1M context| |The lowest price per token|qwen3.8-flash, 1M context| |A mid-price option|qwen3.8-27b| |The Qwen 4 architecture on your own GPUs|Qwen3.8-Flash-Next weights, qwen-community-1.0 licence, 125B total and 6B active| |To switch to Qwen 4 on day one|Keep the model ID in config, so you can swap it the day an ID appears| Weighing other Chinese models while you wait? Z.ai has not announced GLM-5.5 at all, as our [GLM-5.5 release date](https://omniakey.com/blog/glm-5-5-release-date) page shows, and our [Kimi K3.1 Pro review](https://omniakey.com/blog/kimi-k3-1-pro-review) covers a model you can call today. ## Which Qwen models can you call today? qwen3.8-max and qwen3.8-flash are also on OmniaKey's OpenAI-compatible API. See current rates on the [Qwen3.8-Max model page](https://omniakey.com/qwen3-8-max) and the [Qwen3.8-Flash model page](https://omniakey.com/qwen3-8-flash), or on the [pricing page](https://omniakey.com/pricing). ## Frequently asked questions ### When is Qwen 4 coming out? Alibaba's September 22, 2026 press release says Qwen 4 is "currently in training" and gives no date. The same release puts Qwen 4.5 and Qwen 5 at 5 to 10 trillion parameters. As of November 1, 2026, no official catalog lists a Qwen 4 model; the "next week" date comes from third-party sites. ### Will Qwen 4 be open source? Alibaba has said nothing about Qwen 4 weights or a licence. The preview built on the Qwen 4 architecture, Qwen3.8-Flash-Next, ships its weights under the qwen-community-1.0 licence; Alibaba has not said whether Qwen 4 will follow. ### Is Qwen3.8-Flash-Next the same as Qwen 4? No. Qwen's model card, published August 24, 2026, calls it an "experimental preview of the architecture that will underpin Qwen4", with 125B parameters, 6B active and a 262,144-token context. Qwen3.8-Flash is the production model built on it. ### Is there a Qwen 4 API or price yet? There is none today. Model Studio's models page, last updated October 28, 2026, lists qwen3.8-max, qwen3.7-plus and qwen3.8-flash as its current text models, and Hugging Face's Qwen organization and OpenRouter matched it on November 1, 2026. ### What are Qwen 4 Max, Plus, Flash and 27B? Those names come from third-party sites, published between September 22 and October 29, 2026. The September 22 press release names no tiers. Model Studio's current lineup is qwen3.8-max, qwen3.7-plus and qwen3.8-flash. ## Sources checked - [Apsara press release, September 22, 2026](https://www.alibabacloud.com/press-room/alibaba-unveils-roadmap-on-full-stack-ai-strategy) - [Alibaba Cloud blog, September 25, 2026](https://www.alibabacloud.com/blog/alibaba-clouds-2026-apsara-conference-full-stack-ai-roadmap-along-with-global-market-expansion-plan_603598) - [Netherlands region press release, October 29, 2026](https://www.alibabacloud.com/press-room/first-public-cloud-region-in-the-netherlands) - [Amsterdam event page](https://www.alibabacloud.com/en/apsara-conference-amsterdam) - [Qwen3.8-Flash-Next model card](https://huggingface.co/Qwen/Qwen3.8-Flash-Next) - [Qwen organization on Hugging Face](https://huggingface.co/Qwen) - [Model Studio models](https://www.alibabacloud.com/help/en/model-studio/models) - [Model Studio release list](https://www.alibabacloud.com/help/en/model-studio/newly-released-models) - [OpenRouter model list](https://openrouter.ai/api/v1/models) - [QwenLM on GitHub](https://github.com/QwenLM) - [Model Studio billing](https://www.alibabacloud.com/help/en/model-studio/billing-for-model-studio) - [Model Studio text-generation models](https://www.alibabacloud.com/help/en/model-studio/text-generation-model) - [qwen-code pull request 15762](https://github.com/QwenLM/qwen-code/pull/15762) - [transformers pull request 50138](https://github.com/huggingface/transformers/pull/50138) - [llama.cpp pull request 30965](https://github.com/ggml-org/llama.cpp/pull/30965) - [Hacker News comment, October 31, 2026](https://news.ycombinator.com/item?id=50321450) Updated November 1, 2026, 00:21 UTC. We will revise the Qwen 4 status when Alibaba Cloud Model Studio or Qwen's Hugging Face page lists a Qwen 4 model ID or a release date.