TabbitBlog

Qwen 4 Max Preview: How to Access It and What to Expect

Find the exact Qwen4-Max-Preview ID, Token Plan endpoint, limits and end date, and learn how it differs from the separate qwen4-max API model.

In this article
  1. Key takeaways
  2. Qwen4-Max-Preview versus qwen4-max
  3. How to access Qwen4-Max-Preview
  4. Know the shared output limit before setting max tokens
  5. Credits, discounts and actual spend
  6. Decide where the preview belongs
  7. Practical verdict: test the preview, keep a migration plan
  8. Frequently asked questions
  9. What is the exact API model ID?
  10. How do I access Qwen4-Max-Preview?
  11. What is the preview API base URL?
  12. When does the preview end?
  13. Is there a per-token price for the preview?
  14. Can I use it in a production service?

Qwen4-Max-Preview is a temporary Token Plan model, not the same route as the current qwen4-max API model. For the preview, the exact model ID is qwen4-max-preview; Alibaba lists its dedicated API base as https://token-plan.ap-southeast-1.maas.aliyuncs.com. The preview window ends at 24:00 UTC+8 on November 25, 2026, after which Alibaba says the preview model will be retired. Its capabilities may also change during the preview.

That makes this useful for a bounded evaluation, not a model string to pin in a production service. An early Qwen community tester reports using Qoder for a Three.js task and paying in discounted Token Plan credits; the author’s result and cost are one tool-mediated example, not an independent benchmark. The practical question is whether the preview works for your test while it is available—and whether you have a different supported model and endpoint for the workflow that must keep running.

Key takeaways

  • Use the exact preview ID: qwen4-max-preview. Alibaba documents a separate current model, qwen4-max, with a different pay-as-you-go route.

  • Preview access is Token Plan only. Use the dedicated sk-sp- key and the preview’s Singapore-region base URL; standard Model Studio sk- keys are a different billing path.

  • The preview is temporary and may change. Alibaba lists a Nov 11–25 window and says capabilities can be iterated during preview, then the preview is retired.

  • The 393,216-token output budget is shared. Thinking consumes part of the combined limit for thinking and final response.

  • Credits are not tokens or a published USD rate. Preview discounts reduce Credit consumption; Alibaba does not publish a preview per-token USD price.

  • Token Plan has usage rules. Its keys are for interactive coding/agent tools, not custom backends, scripts or non-interactive batch requests.

Qwen4-Max-Preview versus qwen4-max

Detailqwen4-max-previewqwen4-max
StatusTemporary preview; listed Nov 11–25, 2026 (UTC+8)Separately listed as Alibaba Model Studio’s latest model on Nov 16
Access pathToken Plan Personal or Team onlyModel Studio API; documentation lists pay-as-you-go pricing
API key / baseDedicated Token Plan key (sk-sp-); https://token-plan.ap-southeast-1.maas.aliyuncs.comStandard Model Studio route; use the region-specific API endpoint in its guide
Context / output1M context; up to 393,216 output tokens shared by thinking and final answer1M context; up to 131,072 output tokens in non-thinking mode per current model page
Price basisCredits; no published preview per-token USD rateCurrent international price table lists $1.50 input / $4.50 output per million tokens
StabilityCapabilities may be iterated; scheduled retirement after previewSeparate non-preview model entry; still check current docs and account availability

The preview values come from Alibaba Cloud’s Qwen4-Max-Preview page and Token Plan product updates, checked November 16, 2026. The distinct qwen4-max entry is listed in the current model page and pricing table. The $1.50/$4.50 values are the published international rates for the separate model; they are not the preview’s price.

Alibaba’s Nov 16 Token Plan change log lists Qwen4-Max as newly available in both Personal and Team editions. That update does not change the preview ID or make the preview route permanent. If you see both names in a provider, decide based on the exact identifier and endpoint, not the shared “Max” label.

How to access Qwen4-Max-Preview

The official preview page documents the following developer path. This article did not sign into Alibaba Cloud, buy a Token Plan, generate a key or make an API call.

  1. Choose Token Plan Personal or Team. Alibaba’s preview page names those editions as the access route. Availability and subscription details can differ by account.

  2. Generate a dedicated Token Plan key. Token Plan keys start with sk-sp-; standard Model Studio API keys start with sk- and use separate pay-as-you-go billing. Keep the key private and store it in your secret manager.

  3. Use the preview base URL. Set https://token-plan.ap-southeast-1.maas.aliyuncs.com. The ap-southeast-1 route is the dedicated Singapore-region endpoint; use the complete protocol endpoint from Alibaba’s applicable Token Plan guide.

  4. Set the exact model ID. Use qwen4-max-preview. Do not substitute qwen4-max, which is a separate model entry with its own route and price.

  5. Choose a documented protocol and compatible interactive tool. The preview docs list OpenAI Chat Completions, OpenAI Responses and Anthropic Messages compatibility. Token Plan policy allows interactive use within coding and agent tools.

  6. Run a small, reversible evaluation before a longer task. Check the model, endpoint, usage display, and behavior with the same fixed prompt and acceptance check. Do not send sensitive production data to an unverified workflow.

For a broader account setup, the Token Plan overview describes obtaining a dedicated key, while the usage and quota guide explains how credits are deducted. A Token Plan key does not silently fall back to pay-as-you-go when credits run out; Alibaba says calls are blocked once available credits are exhausted.

Know the shared output limit before setting max tokens

The preview has a 1-million-token context and a maximum output of 393,216 tokens, but thinking and the final answer share that output budget. A request with a long reasoning trace can therefore leave less room for the answer than a developer might expect. In OpenAI Chat Completions, Alibaba documents max_tokens as equivalent to max_completion_tokens for this preview: both cap thinking plus the final response.

Thinking is enabled by default. The page lists low, medium and xhigh effort, with xhigh as the default. It also says not to set reasoning_effort and thinking_budget together. When comparing with another model, hold the effort setting and token ceiling constant or label the difference. Otherwise, a longer answer or slower run may reflect configuration, not just model behavior.

This matters for agent loops, too. Save the exact prompt, client version, tool configuration, effort, output cap and timestamp alongside your result. If the preview changes during its window, rerun the same task before concluding that your original observation still holds. Alibaba’s docs explicitly say capabilities continue to be iterated and upgraded during preview.

Credits, discounts and actual spend

Do not confuse Token Plan credits with API tokens. Alibaba describes Credits as the unified unit for the subscription; per-request consumption depends on model unit price, input tokens, cached tokens and output tokens. The preview announcement says consumption is 20% of the standard rate during the preview, with an additional 60% off between 22:00 and 08:00. That is a promotion against standard Credit consumption, not a published dollar-per-token price.

The official preview page does not publish a per-token USD price. The separate Model Studio pricing page lists qwen4-max at $1.50 input and $4.50 output per million tokens in the international table, but those rates apply to qwen4-max, not qwen4-max-preview. Do not use the stable model’s rate to estimate preview Credits.

Token Plan is also a distinct product from ordinary API billing. Alibaba’s permissions and usage policy says it is for interactive use in coding and agent tools. It excludes automation scripts, custom application backends and non-interactive batch calls. A developer should not route a customer-facing product through the preview key just because the key happens to work for a test call.

Decide where the preview belongs

Your situationUse this routeBefore proceeding
You want to explore model behavior in a supported coding/agent toolqwen4-max-preview via Token PlanRecord prompt, effort, tools, date and Credit usage
You need repeatable testing during the preview windowPreview for a short evaluation, with a baseline such as Qwen3.8 MaxRun the same tasks more than once and track the observed model version/date
You are building a backend or unattended batch jobDo not use the preview keySelect an eligible standard API model/plan; confirm its billing and usage terms
You need a route after Nov 25Evaluate the distinct qwen4-max or another current modelChange the model ID, endpoint and price assumptions; rerun acceptance tests
You are choosing between current Qwen modelsCompare the Qwen3.8 Max overview and Qwen3.7 MaxUse your own task set and current provider list

There is no Alibaba-published benchmark table on the preview page. A Reddit user reports one Qoder-based 3D horse task that took about 1 hour 40 minutes and estimates $1.27 without the discount; comments disagree about the output and prompt. Another Reddit thread discusses a filtered answer in the early preview. These are useful signals about evaluation friction, but neither establishes general quality or reliability. A YouTube creator also reports access failures during a four-test run; that account is likewise self-reported. Treat them as reasons to record tool, prompt, date and retry conditions.

If you want to examine related Qwen tooling in a browser, see the Qwen Code browser guide, Qwen3.8 Max browser workflow, Qwen3.7 Max browser workflow, and Qwen3.6 Max Preview browser guide. For a faster prior-generation option, the Qwen3.8 Flash browser guide describes a separate model. Qoder CLI browser setup covers the tool route used in one community example; it does not certify Qwen preview availability.

A temporary preview route crosses a short bridge toward a test workspace, while a separate steady route leads to production servers.
Evaluate a preview in a bounded test; keep the production route and model identifier separate.

Practical verdict: test the preview, keep a migration plan

Qwen4-Max-Preview is worth trying only if the Token Plan route fits your interactive developer workflow and you can finish a useful evaluation before the November 25 retirement. The API identifier, limits and access route are documented; a per-token USD price and a stable future for this preview are not. The model may change during the preview window.

For production or an unattended backend, use a model and billing path whose current terms permit that workload. Alibaba now separately lists qwen4-max as its latest model with a pay-as-you-go API price. That is a different identifier; migrate by changing the configured model and endpoint, then rerun your tests rather than assuming the Preview and current model are drop-in equivalents.

Tabbit can help keep source tabs and notes organized while you compare provider documentation, but this article did not verify that qwen4-max-preview appears in Tabbit’s model picker or call it through Tabbit. If your workflow uses a browser agent, see the Qwen3.8 Max browser-use guide and keep model access, browser access and tool permissions as separate checks.

Frequently asked questions

What is the exact API model ID?

Alibaba Cloud Model Studio lists qwen4-max-preview for the temporary Preview. qwen4-max is a separate current model ID and should not be substituted into preview instructions.

How do I access Qwen4-Max-Preview?

Use Token Plan Personal or Team with a dedicated Token Plan key and the documented Token Plan base URL. The preview is not available through pay-as-you-go billing.

What is the preview API base URL?

The official preview page lists https://token-plan.ap-southeast-1.maas.aliyuncs.com. Use the complete endpoint for the selected API protocol from the current Token Plan guide.

When does the preview end?

Alibaba lists the window as November 11 at 10:00 through November 25 at 24:00, 2026, UTC+8, and says the preview will be retired afterward. Recheck the live docs if your work happens later.

Is there a per-token price for the preview?

No fixed USD-per-token rate is published for the preview. Token Plan bills in Credits, with preview discounts against standard Credit consumption; the separate qwen4-max price is not the preview price.

Can I use it in a production service?

No. It has a scheduled retirement, may change during preview, and Token Plan policy excludes custom backends and non-interactive batch use. Use the preview for a bounded interactive evaluation and choose a permitted current route for production.

FAQ

What is the API model ID for Qwen 4 Max Preview?

Alibaba Cloud Model Studio lists the exact preview ID as `qwen4-max-preview`. The separate current model ID is `qwen4-max`; do not treat the two identifiers as interchangeable.

How do I access Qwen4-Max-Preview?

The preview page says access is through Token Plan Personal or Team, using a dedicated Token Plan API key and base URL. Pay-as-you-go access is not supported for the preview.

What is the Qwen4-Max-Preview API base URL?

The preview documentation lists `https://token-plan.ap-southeast-1.maas.aliyuncs.com` as the dedicated base URL. Use the protocol-specific endpoint from the matching Token Plan guide and confirm the current docs before integration.

How long is the Qwen 4 Max Preview available?

Alibaba lists the preview from November 11, 2026 at 10:00 through November 25, 2026 at 24:00 UTC+8, then says it will be retired. Check the current model page before relying on that endpoint.

How much does the preview API cost per token?

Alibaba does not publish a fixed USD per-token price for the preview; Token Plan usage is measured in Credits and depends on request usage. The preview promotion reduces Credit consumption, but it should not be converted into an invented USD token rate.

Can I use the preview in a production backend?

No. The preview has a scheduled retirement and Alibaba’s Token Plan policy limits use to interactive coding and agent tools, excluding custom application backends, automation scripts and non-interactive batch calls. Evaluate it interactively, then use an eligible non-preview route for production.

Take the next step

Let Tabbit work alongside you.

Research across tabs, automate repetitive browser work, and keep every piece of context within reach.