Qwen3 Max: 256K Context, API & Migration Guide

Qwen3 Max, identified in Alibaba Model Studio as qwen3-max, is a hosted commercial text-generation model with a 256K context window. It is different from qwen3.7-max, which is a separate 1M-context model.

Last verified: July 21, 2026.

Lifecycle notice: Alibaba schedules qwen3-max, qwen3-max-preview, and the listed Qwen3 Max snapshots for deprecation on October 10, 2026 at 00:00:00. The official replacement is qwen3.7-max.

Qwen3 Max key facts

Model type	Hosted commercial text-generation model
Main model ID	`qwen3-max`
Context window	256K tokens
Native input	Text
Output	Text
Public downloadable weights	Not published for this API identifier
Documented replacement	`qwen3.7-max`
Scheduled deprecation	October 10, 2026 at 00:00:00

Qwen3 Max is not Qwen3.7 Max

Feature	`qwen3-max`	`qwen3.7-max`
Context	256K	1M
Provider classification	Legacy/deprecating model	Recommended model
Model ID	`qwen3-max`	`qwen3.7-max`
Interchangeable aliases	No. Applications must explicitly change and test the model ID.

The 1M context specification must not be assigned to qwen3-max. Alibaba’s text-model matrix lists 256K for Qwen3 Max and 1M for Qwen3.7 Max.

For information about the replacement model, see our Qwen3.7 Max guide.

Documented model IDs

qwen3-max
qwen3-max-2026-01-23
qwen3-max-2025-09-23
qwen3-max-preview

When this page was verified, Alibaba’s pricing table mapped the rolling qwen3-max alias to qwen3-max-2026-01-23. A rolling alias can be managed differently from a pinned snapshot, so production migrations should test the exact identifier used by the application.

Thinking, tools, and structured output

Alibaba’s model matrix marks Qwen3 Max as supporting thinking mode, Function Calling, built-in tools, and structured output. Availability and mode behavior can differ by deployment scope and snapshot. For example, some regional tables expose the rolling alias in both thinking and non-thinking modes, while other Global deployments list non-thinking mode only.

Function Calling means the model can return a structured request for an application-defined tool. The surrounding application remains responsible for validating arguments, authorizing the action, executing the tool, and returning the result.

Qwen3 Max pricing

The following standard list prices apply to the Singapore International deployment and were verified against the Alibaba Model Studio pricing document dated July 15, 2026. Prices are per one million tokens and exclude promotional discounts.

Input tokens per request	Input price	Output price
0–32K	$1.20	$6.00
More than 32K–128K	$2.40	$12.00
More than 128K–256K	$3.00	$15.00

Prices vary by deployment scope. The “1 million tokens” shown in the International free-quota column is a conditional usage allowance valid for 90 days after Model Studio activation. It is not the model’s context window.

Node.js API example

Install the OpenAI SDK and set DASHSCOPE_API_KEY and DASHSCOPE_BASE_URL. The base URL must be the OpenAI-compatible Model Studio workspace endpoint for the same region as the API key.

npm install openai

import OpenAI from "openai";

if (!process.env.DASHSCOPE_API_KEY || !process.env.DASHSCOPE_BASE_URL) {
  throw new Error(
    "Set DASHSCOPE_API_KEY and DASHSCOPE_BASE_URL for the same Model Studio region."
  );
}

const client = new OpenAI({
  apiKey: process.env.DASHSCOPE_API_KEY,
  baseURL: process.env.DASHSCOPE_BASE_URL,
});

const response = await client.chat.completions.create({
  model: process.env.QWEN_MODEL || "qwen3-max",
  messages: [
    {
      role: "system",
      content: "Answer accurately and state when evidence is insufficient.",
    },
    {
      role: "user",
      content: "Summarize the main risks in this project proposal.",
    },
  ],
});

console.log(response.choices[0].message.content);

The example remains relevant only while the selected Qwen3 Max identifier is available. Set QWEN_MODEL=qwen3.7-max after validating the replacement in a separate test environment.

Migration checklist

Inventory every application, batch job, evaluation, and prompt that uses a Qwen3 Max identifier.
Test qwen3.7-max with representative production inputs.
Compare instruction following, output structure, tool calls, latency, and token usage.
Revalidate JSON schemas and application-side tool authorization.
Update monitoring, budgets, and model-specific fallback rules.
Switch the production identifier before the documented deprecation timestamp.

Frequently asked questions

Does Qwen3 Max have a 1M context window?

No. Alibaba documents a 256K context window for qwen3-max. The 1M specification belongs to qwen3.7-max.

Can Qwen3 Max be downloaded and self-hosted?

Alibaba presents this identifier as a commercial Model Studio API model. It has not published an official weight download, parameter count, local deployment recipe, or open-weight license for qwen3-max.

Is Qwen3 Max a vision model?

No native visual input is documented for the Qwen3 Max API model. It should be treated as a text-generation model unless an external agent or separately documented visual service is added.

What replaces Qwen3 Max?

Alibaba’s lifecycle documentation identifies qwen3.7-max as the replacement.

Official references

This is an independent technical guide and is not affiliated with or endorsed by Alibaba Cloud or the Qwen team.

Qwen3 Max: 256K Context, API Status, and Migration