The short answer
As of November 12, 2026, Alibaba Cloud’s Model Studio documentation lists Qwen4-Max-Preview, model ID qwen4-max-preview, as a preview available through Token Plan only; pay-as-you-go access is not supported. The documented preview period is November 11, 2026, at 10:00 through November 25, 2026, at 24:00 (UTC+8), after which Alibaba says the preview model will be retired. That end date is not a promised general-release date. (Alibaba Cloud model documentation)
For production planning, treat it as something to evaluate, not depend on: Alibaba documents a scheduled retirement after the preview period, not a successor or uninterrupted service. Alibaba’s September 22 announcement said Qwen 4 was in training; it does not establish the availability, specifications, or commercial terms of the wider Qwen 4 family. (Alibaba Cloud announcement)
What “Qwen 4” refers to
The name is used for different things; keeping them separate prevents mistaken assumptions:
- Qwen 4 is Alibaba’s next-generation model family. Alibaba said on September 22, 2026, that it was in training. (Alibaba Cloud announcement)
- Qwen4-Max-Preview is the preview model Alibaba documents. Its access terms and specifications apply to this model—not automatically to other Qwen 4 models. (Alibaba Cloud model documentation)
- Qwen3.8-Flash-Next is a separate model. Qwen’s official repository describes it as an early preview of architecture intended for Qwen 4; it is not the Qwen 4 family or the Max preview. (Official Qwen repository)
Third-party reports describe a broader set of tiers—Max, Plus, Flash, and 27B. The primary sources cited here establish that Qwen 4 was in training and document Max Preview, but do not confirm current availability or complete specifications for Plus, Flash, or 27B. Don’t use reported tier roles as a basis for choosing a released model.
What the Max preview can do
A November 11 report attributes these capabilities to Alibaba’s documentation for qwen4-max-preview:
- Text, image, and video input.
- Thinking enabled by default. (Yotta Labs report)
The documented context length is 1 million tokens. Maximum output is 393,216 tokens shared between thinking and the final answer. With thinking enabled, reasoning uses part of that output budget; a limit carried over from an answer-only setup may therefore truncate the response sooner than expected. (Alibaba Cloud model documentation)
These reported capabilities are useful for initial screening, not a guarantee of quality or latency on a particular task. The documentation cited here does not provide comparative benchmark results. A November 11 third-party report likewise said published Max Preview benchmarks were not available at that time; treat community scores or demos as preliminary unless their model, test, and method are clearly documented. (Yotta Labs report)
API access, billing, and the production question
Alibaba says the preview is available through Token Plan only; pay-as-you-go access is not supported. Alibaba’s Token Plan overview describes the service as billed in Credits for interactive use in compatible AI tools. That describes the preview access and billing channel; it does not establish a separately priced, stable pay-as-you-go production API. (Alibaba Cloud model guide; Token Plan overview)
The source material cited here does not establish the plans’ dollar prices, whether a particular plan or account is eligible for this preview, or preview-specific credential setup. The model-specific per-token price is also not established in these materials. Check the current plan price, preview eligibility, and model allowance in your Alibaba Cloud console before subscribing; don’t treat Credit billing as a published per-token rate. (Token Plan overview)
Check regional setup before relying on access. Alibaba’s English Personal Edition overview says that edition is available only in Singapore, while its Chinese Qwen4-Max-Preview guide shows a Beijing-region endpoint. These are region-specific documentation paths; they do not establish eligibility for a particular account or US availability. Confirm the supported region and endpoint shown in your Model Studio console. (English Personal Edition overview; Chinese Qwen4-Max-Preview guide)
Production decision: The preview may be worth a bounded evaluation if your account can access it and your task benefits from its reported multimodal or long-context capabilities. But a model scheduled for retirement on November 25 is a poor dependency for a production launch that must keep working beyond that date. The retirement date does not promise a successor, general release, or uninterrupted API availability. (Alibaba Cloud model documentation)
What remains unknown—and what not to assume
The official materials cited here do not establish a general release date for the full Qwen 4 family, or current availability and complete specifications for Plus, Flash, or 27B. They also do not establish post-preview billing or long-term service terms. Alibaba’s September announcement that Qwen 4 was in training is not a release schedule. (Alibaba Cloud announcement)
Keep the architecture preview separate from the Max preview. Qwen describes Qwen3.8-Flash-Next as an early look at architecture intended for Qwen 4, but its design details and performance results are not specifications for Max, Plus, Flash, or 27B. Likewise, do not transfer predecessor-model hardware requirements, licenses, or prices to Qwen 4. (Official Qwen repository)
How to evaluate it before choosing
If you can access the preview, run a small, task-specific comparison rather than relying on a model label or a single demo:
- Select representative prompts from your real workload, including difficult and failure-prone cases.
- Compare outputs against your current model using the same inputs and success criteria. For tool use, record whether the correct tool was called and the task completed—not just whether the answer sounded plausible.
- Measure the operational factors that matter to you: successful task rate, output quality, latency, and Token Plan Credits consumed. Don’t convert the plan price into an assumed per-token rate without actual usage data.
- Test the output budget with thinking enabled. Alibaba documents that thinking and the final answer share the maximum output allowance; leave room for both.
- Treat the result as a preview evaluation only. Before production use, confirm that the model and billing path will remain available for your required period, and review the applicable service terms.
Bottom line
Alibaba documents Qwen4-Max-Preview for Token Plan access, but it is a time-limited preview, not a stable pay-as-you-go production choice. Its reported multimodal and long-context capabilities make it a candidate for short, controlled testing—not proof of workload fit or ongoing availability. If you need a production dependency today, wait for confirmed post-preview terms and evaluate it against an available option. If you can access the preview, test it without making your launch depend on it.
Sources
Latest model: Qwen4-Max-Preview — Alibaba Cloud Model Studio
Alibaba Unveils Roadmap on Full-Stack AI Strategy — Alibaba Cloud
Qwen3.8-Flash-Next — Official Qwen GitHub repository
Token Plan overview — Alibaba Cloud Model Studio
Token Plan (Personal Edition) — Alibaba Cloud Model Studio
Qwen4-Max-Preview — Alibaba Cloud Model Studio (Chinese)
Qwen 4: Release Date, Qwen4-Max Preview, and What's Confirmed (2026) — Yotta Labs