When Is Qwen4 Coming Out? The Max Preview and Token Plan

In short: Alibaba has not said when the Qwen4 family will be generally available. As of November 12, 2026, the only model you can use is a preview of Qwen4-Max. It runs only in the cloud, through Alibaba's Token Plan subscription, and the preview closes at 7:00 p.m. Istanbul time on November 25, 2026. No open weights, model card or license have been published, so you cannot install Qwen4 on your own computer or server today.

Alibaba said in September that Qwen4 was still in training. On November 11, 2026, the first model of the family opened, but as a time-limited preview rather than a general release. So the question of when Qwen4 is coming out has two parts: the Max preview, which stays open for two weeks, and the date the family reaches everyone, which has not been announced. The only way into the preview is a subscription called Token Plan, and its terms decide where and how you can use the model.

Qwen4 is the next generation of Alibaba's Qwen model family, and its only member you can use today is a preview version of Qwen4-Max. In its 2026 Apsara Conference announcement, published on September 25, 2026, Alibaba Group said its next-generation model, Qwen4, was in training.

The model that arrived in November is Qwen4-Max-Preview. Alibaba Cloud Model Studio's Qwen4-Max-Preview page introduces it as the next-generation flagship multimodal model in the Qwen family. The November 11, 2026 entry in Qwen Cloud's model changelog says the model is in preview and available through Token Plan only. According to the same entry, it combines text generation, image and video understanding and deep reasoning, and it is designed for coding, office work, frontend design and complex agent tasks.

Qwen4 is the family name, and the model ID you select in tools is qwen4-max-preview. Qwen Cloud's Token Plan Team Edition documentation says the model also has an alias, qwen4.0-max-preview, and that bills show that name.

There is no official release date for the Qwen4 family. Alibaba said in September that the model was in training and opened the Max preview in November, but its official pages do not say when the family will become generally available. As of November 12, 2026, the only dated item is the Qwen4-Max preview window.

According to the Model Studio page, the model will not change during the preview and will be retired when the preview ends. Alibaba has not said what replaces it. If you plan to use Qwen4-Max after November 25, check Alibaba's current documentation first.

As of November 12, 2026, the one certain fact about Qwen4 is that the Max preview is open on Token Plan. A release date for the family, pay-as-you-go API access and downloadable weights do not exist yet.

To try Qwen4-Max, you buy a Token Plan subscription, create an API key for the plan and enter that key in a supported coding or agent tool. Qwen Cloud's Token Plan Overview describes Token Plan as a subscription billed in Credits that lets you use several models with one plan. According to the same page, Token Plan keys start with sk-sp- and cannot be used in place of pay-as-you-go keys, or the other way around.

Qwen Cloud's Latest model page says both the individual and the team editions of Token Plan give access to Qwen4-Max, while pay-as-you-go access is not supported. Once the key is in your tool, you select qwen4-max-preview as the model.

Token Plan is built for interactive use in coding and agent tools such as Claude Code, Cursor, Qwen Code, Qoder and OpenClaw. RuneLab's Agentic Coding for Software Engineering Teams course teaches how to hand multi-file tasks to an agent in these kinds of tools, with hands-on work in Claude Code, Cursor Agent and OpenAI Codex CLI. It also covers Plan-Act-Verify loops and tool integration through MCP.

Personal Edition, the individual version of Token Plan, comes in four plans. As of November 12, 2026, Alibaba Cloud's Token Plan Personal Edition Overview lists these prices and quotas:

These quotas are new. According to Qwen Cloud's November 9 price update, all four quotas went up on November 9, 2026, and the Pro plan dropped from $68 to $43. The new-user price on Lite and Essential runs until 11:59:59 p.m. on November 30, 2026 (UTC+8), which is 6:59:59 p.m. in Istanbul, and only applies to accounts that have never bought Personal Edition.

If the monthly quota runs out, an Extra Bundle adds 20,000 Credits for $15. You need an active subscription to buy one, you can hold up to five at a time, and once the monthly quota is used up, Credits come out of the bundle automatically. Each bundle lasts one month.

Two rules matter if you want to stop. Personal Edition cannot be canceled at the moment, but according to Qwen Cloud's Token Plan Personal FAQ, you can turn off auto-renewal on the Token Plan page of the console. The same page says the plan is non-refundable.

Qwen4-Max runs at a discount during the preview. According to the Token Plan Personal Edition Overview, Qwen4-Max calls use as little as 20% of the standard rate in Credits, which gives five times the usage from the same quota. Calls between 10:00 p.m. and 8:00 a.m. UTC+8, or 5:00 p.m. to 3:00 a.m. in Istanbul, get an extra 60% off, which brings usage down to 8% of the standard rate. Alibaba reserves the right to change these promotions.

The documentation gives no fixed Credits figure per call for Qwen4-Max, because usage depends on the number of tokens processed, the thinking mode and tool calls. In most models, Turkish text takes more tokens than English text with the same content, so a team working in Turkish can use up its quota faster. We explain why in What is a token in AI?

No. As of November 12, 2026, Qwen4's open weights, model card and license have not been published. Neither Qwen's Hugging Face account nor the Ollama library has a model named Qwen4, and the research page where the team announces its models has no Qwen4 post. So there is no official ollama pull command, GGUF file or VRAM table either. Do not trust download links shared under the Qwen4 name unless they come from Qwen's official accounts.

If you need a Qwen model running on your own server today, the route is the previous generation. Our guide to running an LLM locally covers how to install open-weight Qwen3 models and how much GPU memory they need. RuneLab's Hati Glint SQL comes from this generation too: an open model fine-tuned from Qwen3.5-9B, released under Apache 2.0 on Hugging Face, that turns Turkish questions into SQL and can call tools.

If an open-weight Qwen4 comes out, the first step is reading its license, because the license decides whether the model allows commercial use and on what terms. We compare open model licenses in Open-weight model licenses and commercial use. If the license works for you, the next question is whether the model fits your work. RuneLab's Open-Weight LLM Selection, Licensing and Benchmarking for Enterprise Use course covers choosing open models such as Llama, Qwen, Mistral, Gemma and DeepSeek by license terms, the trade-offs between dense and MoE architectures, Turkish performance and tool calling.

Token Plan is designed for employees' personal, interactive use in coding tools. It is not an API for a company's own applications, and the data you send is processed globally from the Singapore region. Those two conditions decide what a company can try Qwen4-Max for.

According to the Token Plan Personal Edition Overview, the plan is for interactive use in coding and agent tools only. Automation scripts, custom application backends and non-interactive batch calls are out of scope. Out-of-scope use counts as a violation and can lead to the subscription being suspended or the API key being banned. The key is for the subscriber's own use and cannot be shared.

In practice, this means you cannot put a Token Plan key into a company application, an automation or the backend of a customer chatbot. There is also no other way today to call Qwen4-Max from your own application through an API, because the model is not available on pay-as-you-go.

For team use, Qwen Cloud's Token Plan Team Edition documentation offers three seat types, priced per seat per month: Standard at $20 (25,000 Credits), Pro at $75 (100,000 Credits) and Max at $180 (250,000 Credits). Qwen4-Max is on this edition's model list too. According to the documentation, Team Edition does not use conversation data for model training. The usage rule is the same as for the individual edition: interactive use in compatible tools only, with no automated scripts or application backends.

According to Alibaba's Token Plan Personal Edition Overview, Token Plan's only region is Singapore, its only deployment mode is Global, and model inference is performed globally. The documentation says prompts and model outputs involve cross-border data transfer and that making sure your use of Token Plan complies with applicable laws is the user's responsibility. The Team Edition documentation describes the same region and the same transfer.

None of this makes trying Qwen4-Max a breach of KVKK, Turkey's personal data protection law. Using a cloud tool is not a KVKK breach on its own, and running a model in-house does not make you compliant on its own. What decides it is which data you send to the tool and on what basis that data goes abroad. So before a trial, settle two questions with your legal team: will the code or documents you send to the tool contain personal data, and if so, on what basis will that data be transferred abroad? We cover what to check before entering personal data into generative AI tools in our KVKK generative AI guide, and which work goes to the cloud and which stays in-house in Local LLM vs ChatGPT.

If data must never leave the company, Qwen4-Max cannot meet that need today, because it runs only in Alibaba's cloud. Companies in that position run open-weight models on their own network. Weighing which work belongs on a cloud API and which stays in-house, on cost, risk and regulation, is what RuneLab's Cloud LLM APIs vs. On-Premise Models: Cost, Risk and Regulation course works through. For teams that choose in-house, RuneLab's RuneBox is on-premise LLM infrastructure installed on the company network: models, RAG and agent workflows run on a box with NVIDIA GPUs. Because data is processed within the organization's boundaries, it makes your BDDK and KVKK assessments easier. Compliance remains your organization's own assessment.

Qwen4 is the next generation of Alibaba's Qwen model family. Its only member available today is a preview of Qwen4-Max, which runs in the cloud and only through a Token Plan subscription. There is no downloadable version.

There is no official release date for the Qwen4 family, and Alibaba said in September that the model was in training. As of November 12, 2026, only the Qwen4-Max preview exists, and it is open from 5:00 a.m. on November 11 to 7:00 p.m. on November 25, Istanbul time.

Yes, Qwen4-Max is available only through the paid Token Plan subscription. Token Plan is Alibaba's subscription billed in Credits that covers several models. As of November 12, 2026, the individual plans cost $6, $10, $18 and $43 a month, and new users pay half price on the first two.

No. As of November 12, 2026, Qwen4's open weights, model card and license have not been published, so there is no official Ollama or local install path. If you need a Qwen model that runs locally, the option today is the open-weight Qwen3 models.

The preview ends at midnight (24:00) on November 25, 2026, UTC+8, which is 7:00 p.m. in Istanbul. According to Alibaba's documentation, the model does not change during the preview and is retired when the preview ends.

There is no way today to call it from your own application through an API, because Qwen4-Max is not available on pay-as-you-go. A Token Plan key only allows interactive use in tools such as Claude Code, Cursor or Qwen Code and cannot be used in automation scripts or application backends.

According to Alibaba's documentation, Token Plan's only region is Singapore, its deployment mode is Global and inference is performed globally. The documentation says prompts and model outputs involve cross-border data transfer and that legal compliance is the user's responsibility.

• Alibaba Group, Alibaba Cloud's 2026 Apsara Conference: Qwen4 being in training
• Alibaba Cloud Model Studio, Qwen4-Max-Preview: the model's description, the preview times, no changes during the preview and retirement after it, Token Plan-only access
• Qwen Cloud, Latest model: the model ID, access through the individual and team editions, no pay-as-you-go access
• Qwen Cloud, Model changelog: the November 11, 2026 entry, preview status and the model's capabilities
• Alibaba Cloud Model Studio, Token Plan Personal Edition Overview: plan prices and quotas, the new-user price, Extra Bundles, cancellation, scope of use, supported tools, region and cross-border transfer, the Qwen4-Max discount
• Qwen Cloud, Token Plan Individual price reduction notice: the quota increase and Pro price of November 9, 2026
• Qwen Cloud, Token Plan Overview: what Token Plan is, the plan-specific key and the sk-sp- prefix
• Qwen Cloud, Token Plan Personal FAQ: turning off auto-renewal, refunds and key sharing
• Qwen Cloud, Token Plan Team Edition Overview: seat prices, training data, scope of use, region, model list and alias
• Hugging Face, Qwen: the open-weight models Qwen has published
• Qwen, Research: the Qwen team's model announcements

• How to run an LLM locally: offline AI setup guide
• Local LLM vs ChatGPT: how to decide under KVKK
• KVKK and generative AI guide
• Open-weight model licenses and commercial use
• What is a token in AI? Why Turkish text uses more tokens
• What is an LLM? How large language models work
• Courses: Open-Weight LLM Selection, Licensing and Benchmarking, Agentic Coding for Software Engineering Teams