Free guide

Qwen4-Max Preview: What It Is and How to Try It Before November 25

Alibaba's next flagship is out as a two-week preview. What it is, what people built with it on day one, how to set it up in Claude Code, and the catches.

By Beau Johnson·
The Short this guide goes with: Qwen4-Max Preview: What It Is and How to Try It Before November 25

Quick answer

Qwen4-Max Preview (model ID qwen4-max-preview) is Alibaba's next-generation flagship Qwen model. It reads text, images and video, has a 1M-token context window, and is built for coding, front-end design and agent work. You can only use it through Alibaba's paid Token Plan (or early access in Qoder, QwenWork, the Qwen app and QwenStudio) until the preview is retired at the end of November 25, 2026 (UTC+8).

The short version

  • Qwen4-Max Preview is Alibaba's next flagship: text, image and video in, text out, 1M tokens of context, thinking on by default.
  • It's a two-week preview: Token Plan only (no pay-as-you-go), November 11 to 24:00 on November 25, 2026 (UTC+8), then it's retired.
  • Alibaba published no benchmarks and no per-token price. It asked people to test it on real problems, so early builds are the only evidence so far.

How it actually works

Think of it as a loaner car

You don't buy Qwen4-Max Preview by the token. You get the keys for a fixed window through a subscription, drive it on your own work, and hand it back. Alibaba says no changes will be made during the preview, and the model is retired when it ends.

One model, three kinds of input

It takes text, images and video as input and answers in text, so it can look at a screenshot, a document or a video clip. The context window is 1 million tokens, and one answer can run up to 393,216 tokens, shared between its thinking and the final reply.

It thinks first, and you pick how hard

Thinking is on by default. The reasoning_effort setting takes low, medium or xhigh (the default). Alibaba says not to set reasoning_effort and thinking_budget together.

It speaks OpenAI and Anthropic

The endpoint supports OpenAI Chat Completions, the Responses API and Anthropic Messages, plus function calling, web search and JSON Schema output. That's why tools like Claude Code, Codex, Qwen Code and OpenClaw can drive it with a new base URL and key.

How to set it up

  1. 1.Try it with no setup

    Early access is open in Qoder (on every Qoder product and plan, per Qoder), QwenWork, the Qwen app and QwenStudio. Pick Qwen4-Max-Preview in the model list.

  2. 2.Or subscribe to a Token Plan Individual plan

    Plans are Lite ($6 a month, $3 for new users), Essential ($10, $5 for new users), Standard ($18) and Pro ($43). The new-user price runs until November 30, 2026 (UTC+8). Then copy your Token Plan API key from home.qwencloud.com/api-keys.

  3. 3.Point Claude Code at it

    Add these to ~/.claude/settings.json under "env" (or export them in one terminal). This maps Claude Code's Opus slot to Qwen4-Max Preview. The base URL must end in /apps/anthropic with no /v1.

    ANTHROPIC_AUTH_TOKEN="YOUR_TOKEN_PLAN_KEY"
    ANTHROPIC_BASE_URL="https://token-plan.maas.qwencloudapi.com/apps/anthropic"
    ANTHROPIC_DEFAULT_OPUS_MODEL="qwen4-max-preview"
  4. 4.Check it works, then test it on real work

    Open a new terminal and run the line below. If long answers get cut off, set CLAUDE_CODE_MAX_OUTPUT_TOKENS=64000. Then give it a task you already know the answer to, from your own repo, and compare it with the model you use now.

    claude -p "Hello"

The fine print

  • Preview only: November 11, 10:00 to November 25, 24:00, 2026 (UTC+8), which is 8 AM Pacific on November 25. The preview model is retired after that.
  • Token Plan only. Pay-as-you-go keys don't work with it, and mixing a Token Plan key with another base URL gives a 401 invalid_api_key error.
  • The Individual plan is for interactive use in coding and agent tools. Alibaba says it must not be used for automation scripts, app backends or batch jobs, and keys can't be shared.
  • Singapore region only, so your prompts and outputs cross borders.
  • No benchmarks, per-token price or open weights have been published for this preview. Early reviews on X are mixed: one builder called its 3D work so-so.

FAQ

What is Qwen4-Max Preview?

It's the preview of Alibaba's next-generation flagship Qwen model, model ID qwen4-max-preview. It reads text, images and video, has a 1M-token context window and is built for coding, front-end design and agent work.

How long is Qwen4-Max Preview available?

From 10:00 on November 11 to 24:00 on November 25, 2026 (UTC+8). Alibaba says the preview model will be retired after that. The wider Qwen4 family is described as coming.

Is Qwen4-Max free?

No. Through the API it's only on Alibaba's paid Token Plan, which starts at $6 a month ($3 for new users until November 30). Qoder offers it inside its own plans.

Can I use Qwen4-Max in Claude Code?

Yes, with a Token Plan key. Set ANTHROPIC_BASE_URL to https://token-plan.maas.qwencloudapi.com/apps/anthropic, your Token Plan key as ANTHROPIC_AUTH_TOKEN, and qwen4-max-preview as the Opus model.

Are Qwen4-Max weights open?

Not for this preview. It's only served through Alibaba's Token Plan and partner apps, and no weights have been published.

Sources

Facts checked 2026-11-11 against QwenCloud's latest-model page, the Token Plan Individual overview, the Claude Code setup page, and the official Qwen and Qoder posts on X.

Want help building with tools like this?

Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.

Join Shipping Skool

More free guides