Who this is for
This guide is for developers who want to try Qwen4-Max-Preview from an API or a coding tool. By the end you should know how the model is reached today, which settings to make, and how thinking mode and the output limit affect results.
What Qwen4-Max-Preview is
Qwen4-Max-Preview is the next-generation flagship multimodal model in the Qwen family. It combines text generation, image and video understanding and deep reasoning, and it is designed for coding, office productivity, frontend design and complex agentic tasks[1]. The model ID is qwen4-max-preview[1].
The model accepts text, images and video for screenshot analysis, document understanding and video analysis[1]. Its context length is 1M tokens. The maximum output is 393,216 tokens, shared by thinking and the final answer[1]. Tool support covers Function Calling, web search and built-in tools. Structured outputs support JSON Object and JSON Schema[1].
Access: Token Plan only, for now
Qwen4-Max-Preview is in preview and is available only through Token Plan. Pay-as-you-go access is not supported[1]. The preview runs from 10:00 on November 11, 2026 to 24:00 on November 25, 2026 (UTC+8). No changes will be made during the preview, and the model will be retired when it ends[1]. It suits trials and evaluation, not a long-term production dependency.
Token Plan is a subscription service from Alibaba Cloud Model Studio. It is billed in Credits and is meant for interactive use in compatible AI tools[2]. It comes in a Personal edition and a Team edition, and each has its own configuration guide[2].
Get a dedicated API key
Choose the Personal or Team edition, follow its guide to subscribe, and obtain a dedicated API key[1]. The Token Plan quick start links to console pages for buying a plan and getting an API key, in the ap-southeast-1 region[2]. The examples show Token Plan keys with the sk-sp- prefix, written as sk-sp-xxxx placeholders[2].
Call the model
- Set the Base URL to
https://token-plan.ap-southeast-1.maas.aliyuncs.com. For the complete endpoint of each protocol, follow the Token Plan guide[1]. - Set the model ID to
qwen4-max-preview[1]. - Choose the interface that matches your protocol. The model is compatible with the OpenAI Chat Completions, Responses and Anthropic Messages APIs[1].
If you use a coding tool, the Token Plan overview's quick start configures the agent from the command line. The Claude Code example is[2]:
curl -fsSL https://bailian.aliyun.com/cli/install.sh | bash
export TOKEN_PLAN_API_KEY="sk-sp-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
bl config agent --agent claude-code --region ap-southeast-1 --api-key $TOKEN_PLAN_API_KEY --model autoThe overview lists compatible tools including Claude Code, Codex, Qwen Code, OpenClaw, OpenCode and Trae[2].
Thinking and the output limit
Thinking is enabled by default. reasoning_effort adjusts how hard the model reasons, with the options low, medium and xhigh, and xhigh is the default[1]. Do not set reasoning_effort and thinking_budget together[1].
The output limit is the easiest place to get caught. For OpenAI Chat Completions, max_tokens is equivalent to max_completion_tokens for qwen4-max-preview. Both limit the combined tokens for thinking and the final answer. Thinking uses the same budget, so a limit sized only for the answer can truncate the output. Existing Qwen models keep their previous behaviour[1].
In multi-turn chats, Chat Completions enables preserve_thinking by default. Return all historical thinking in the reasoning_content field of assistant messages, not concatenated into content[1].
Tools and multimodal input
Function Calling, web search and built-in tools are supported. The page links to separate guides for tool definitions and workflows, for visual input (Image and video understanding), and for output formats (Structured outputs)[1].
The Token Plan overview notes that AI tools natively handle text only. Image, video and speech capabilities are connected through extensions[2].
Common questions
- Calls fail: check that you are using a Token Plan key and the Token Plan Base URL. The model is offered only through Token Plan and is not sold pay-as-you-go[1].
- Answers are cut short: check whether
max_tokensalso has to cover the thinking tokens, and raise the limit if needed[1]. - After the preview: the model will be retired, so keep the model ID in configuration where you can change it[1].
Next steps
The page links to the separate guides for visual input, function calling and structured outputs. For client-specific setup, follow the Token Plan guide for your tool[1][2].
Sources
- 1Qwen4-Max-Preview (Alibaba Cloud Model Studio)Alibaba Cloud
- 2Token Plan overviewAlibaba Cloud Model Studio docs
Cite this page: Calling Qwen4-Max-Preview through Token Plan. TellWorld, https://www.tellworld.net/practice/practice-qwen4-max-preview-token-plan. Updated 2026-11-15