By Jonathan West · Updated Nov 17, 2026

Qwen4-Max-Preview Explained

What Alibaba documents about the preview and what a business can test before it closes.

By Jonathan West · Updated Nov 17, 2026

Yes, Qwen4-Max-Preview is currently available for testing. Alibaba Cloud launched the preview on November 11, 2026, as a multimodal artificial intelligence model that accepts text, image, and video inputs. Access is restricted exclusively to Alibaba Cloud Token Plan subscribers, and the evaluation window closes at 24:00 UTC+8 (Coordinated Universal Time plus eight hours) on November 25, 2026.

The model features a context length of 1M tokens and a maximum output ceiling of 393,216 tokens. That output limit is shared dynamically between intermediate thinking tokens and the final answer generation. Alibaba Cloud provides compatibility with OpenAI Chat Completions, Responses, and Anthropic Messages protocols, alongside native support for Function Calling, web search, and structured JavaScript Object Notation (JSON) schemas.

Alibaba Cloud will retire the model after the preview ends.


Qwen4-Max Is Out of Preview

Yes. Alibaba Cloud's Qwen4-Max documentation says the production model is available and recommended, at $1.50 per million input tokens and $4.50 per million output tokens in Singapore. The qwen4-max-preview ID retires after 24:00 UTC+8 on November 25, 2026. Artificial Analysis reports an Artificial Analysis Intelligence Index v5.0.1 score of 43 (checked November 16, 2026).

If you came to this page for the preview, use the production model ID when planning access beyond that retirement date. The two IDs have different output limits and reasoning defaults. For the wider Qwen4 release, see what Qwen 4 is and which models are out.

Itemqwen4-max-previewqwen4-max (production)
StatusPreview from 10:00 November 11 through 24:00 November 25, 2026 (UTC+8)Available and recommended by Alibaba Cloud
AccessToken Plan only. Pay-as-you-go not supportedPay as you go through Alibaba Cloud Model Studio, model ID qwen4-max
PriceToken Plan only. Pay-as-you-go not supportedSingapore: $1.50 input and $4.50 output per million tokens. Beijing, Hong Kong, Frankfurt, Virginia and Tokyo: $1.413 input and $4.24 output
RetirementAfter 24:00 UTC+8 on November 25, 2026No retirement date listed
Context1,000,000 tokens1,000,000 tokens
reasoning_effortlow, medium, xhigh (default xhigh)low, high, xhigh (default high)
Maximum output393,216 tokens shared by thinking and final answer131,072 non-thinking tokens. 262,144 tokens combined for thinking and final answer

The context window stays at one million tokens, but do not carry the preview's output setting or default reasoning level into a production request without reviewing it. Alibaba Cloud says that, starting with qwen4-max-preview, max_tokens equals max_completion_tokens and limits thinking plus the answer together. Existing Qwen models retain their previous behavior. For production, Alibaba Cloud also maps minimal to low, medium to high and max to xhigh.

Almost nobody should choose the preview for work that must continue beyond November 25. It has a stated retirement time and requires Token Plan access. If you need to price an invocation per million tokens, use the production figures for your region rather than assigning the preview a pay-as-you-go rate. The production price shown for Singapore differs from the price Alibaba Cloud lists for its other named regions.

This recommendation would change if Alibaba Cloud extended the preview or changed its production recommendation. For a production request, compare the regional rates in Qwen4-Max pricing and the model choice in Qwen4-Max vs Qwen3.8 Max before selecting an ID.


Qwen4-Max-Preview at a Glance

Qwen4-Max-Preview operates as the next-generation flagship multimodal model in the Qwen family, uniting text generation, image and video understanding, and deep reasoning under the model identifier qwen4-max-preview. The system runs exclusively during a fixed evaluation window from 10:00 on November 11, 2026, through 24:00 on November 25, 2026 (UTC+8). It requires an active Token Plan Personal or Team tier key, with standard pay-as-you-go billing explicitly unsupported.

Specification ItemWhat Alibaba Documents
Model Identifierqwen4-max-preview
Release StatusPreview (retired after the window closes)
Evaluation WindowNovember 11, 2026 (10:00) to November 25, 2026 (24:00 UTC+8)
Billing and AccessToken Plan Personal or Team key only (no pay-as-you-go)
Supported Input ModalitiesText, image, and video
Context Window Capacity1M tokens
Maximum Output Ceiling393,216 tokens (combined thinking and visible answer)
Tool IntegrationFunction Calling, web search, and built-in tools
Structured Data OutputJSON Object and JSON Schema
Reasoning EngineThinking enabled by default with three effort levels
Protocol CompatibilityOpenAI Chat Completions, Responses, and Anthropic Messages APIs

These technical specifications come directly from official vendor documentation published on the Alibaba Cloud Model Studio qwen4-max-preview page, last updated on November 16, 2026.


How to Get Access Before the Preview Closes

Access to the qwen4-max-preview endpoint requires a dedicated Application Programming Interface (API) credential generated under Alibaba Cloud Model Studio Token Plan Personal or Token Plan Team subscriptions. Pay-as-you-go access is not supported. The documentation says model capabilities continue to be iterated and upgraded during the preview, and Alibaba Cloud will retire the model identifier once the preview finishes at 24:00 on November 25, 2026 (UTC+8).

Alibaba's Token Plan guides cover supported clients, configuration steps and Credits benefits. Alibaba gives a dedicated base URL for the Token Plan, with the host token-plan.ap-southeast-1.maas.aliyuncs.com.


Documented Capabilities of Qwen4-Max-Preview

Alibaba Cloud says Qwen4-Max-Preview is designed for coding, office productivity, frontend design and complex agentic tasks. It accepts text, images, and video for screenshot analysis, document understanding, and video analysis.

Tool calling supports Function Calling, web search, and built-in tools. Structured outputs support JSON Object and JSON Schema. The model is compatible with OpenAI Chat Completions, Responses, and Anthropic Messages APIs.

Alibaba lists these reasoning and output rules:

  • Thinking is enabled by default and shares the output budget with the final answer.
  • Reasoning intensity is controlled using the reasoning_effort parameter, accepting values of low, medium, or xhigh (xhigh is the default).
  • Do not set reasoning_effort and thinking_budget together.
  • In OpenAI Chat Completions, max_tokens is equivalent to max_completion_tokens, limiting the combined tokens for thinking and the final answer.
  • Thinking consumes part of the same budget, so a limit sized for the answer alone may truncate output.
  • Multi-turn interactions in Chat Completions activate preserve_thinking by default, requiring client applications to return all historical reasoning blocks inside the reasoning_content field of the assistant message rather than concatenating it into standard message content.

Metrics and Operational Details Alibaba Does Not Publish

The preview page states no price. Qwen4-Max has published rates: see Qwen4-Max pricing. For other models with published prices, see AI model pricing.

The page also states no benchmark results and no rate limits. It states no parameter count and no open-weights status. Infrastructure routing is documented only for the Singapore regional host (ap-southeast-1), with no other regions stated.

Alibaba states no rate limits, so measure throughput and latency on your own task during the preview window.


Qwen4-Max-Preview Compared with Qwen 3.6

Alibaba Cloud publishes no head-to-head comparison or benchmark between Qwen4-Max-Preview and Qwen 3.6.

For the production model against its predecessor, see Qwen4-Max vs Qwen3.8 Max.

When this page was first published on November 13, 2026, the official Alibaba Cloud Model Studio models directory listed qwen3.7-plus, qwen3.8-flash, and qwen4-max-preview, but not Qwen 3.6. Because Alibaba Cloud does not list Qwen 3.6, no Qwen 3.6 specification from Alibaba is available to compare. Developers looking for setup instructions for a newer Qwen model can read how to use Qwen 3.8 Max for practical implementation guidance.

Evaluation QuestionQwen4-Max-PreviewQwen 3.6
Published Commercial PriceNot stated by Alibaba CloudNot verified on Model Studio directory
Vendor Head-to-Head BenchmarksNone publishedNone published
Endpoint Access MethodToken Plan subscription through November 25, 2026Not listed on active model directory

What Businesses Can Test Before November 25

Teams with Token Plan access can run short checks before the preview window closes.

These checks reason from Alibaba's documented capabilities and are not test results. Run them on your own tasks:

  • Long-document parsing: Feed multi-hundred-thousand-token technical manuals, compliance audits, or code repositories to assess retrieval quality across the 1M-token context buffer.
  • Structured data generation: Supply nested JSON Schema definitions to verify whether output strictly adheres to required keys, types, and formatting constraints under complex extraction loads.
  • Tool invocation stability: Test multi-step Function Calling sequences combined with web search retrieval to evaluate autonomous agent reliability.
  • Visual document extraction: Submit high-resolution architectural schematics, interface screenshots, or brief operational video sequences to evaluate multimodal comprehension.
  • Reasoning token allocation: Set reasoning_effort to low, medium, and xhigh while observing output token consumption to determine how intermediate thinking impacts downstream task budgets.

Deployment Boundaries

Qwen4-Max-Preview is not suitable for organizations seeking permanent production backends, pay-as-you-go billing, or a published price. Because Alibaba Cloud will retire this model version at the conclusion of the evaluation period, a live customer-facing workflow built on this endpoint has no model to call after the preview ends.

For models whose weights can be downloaded, see the guide to the best open-weights AI models. This would change if Alibaba Cloud extends the preview past November 25, 2026.

To evaluate this preview effectively, select one complex document processing workload, run the queries against qwen4-max-preview using a valid Token Plan key before November 25, 2026, and record the output fidelity before the endpoint shuts down.

Frequently Asked Questions

  • Yes. Alibaba Cloud launched Qwen4-Max-Preview on November 11, 2026. It remains accessible for testing until 24:00 on November 25, 2026 (UTC+8), after which the preview model will be retired.
  • Alibaba Cloud restricts access exclusively to Token Plan Personal or Token Plan Team subscribers and does not publish a standard per-token price. Alibaba Cloud does not state whether use consumes plan credits or incurs additional fees.
  • No. The official Model Studio documentation explicitly states that pay-as-you-go billing is not supported for this preview. Access requires an active Token Plan subscription key.
  • The model will be retired after the preview ends at 24:00 UTC+8 on November 25, 2026. Alibaba Cloud says capabilities continue to be iterated and upgraded during the preview.
  • The Alibaba Cloud Model Studio page does not state whether Qwen4-Max-Preview will be released as open weights. It lists the model as available through Token Plan.

Plan a Qwen4-Max-Preview Evaluation

Book a consultation with Layer3Labs to plan an evaluation of Qwen4-Max-Preview before November 25, 2026.

Book a Consultation