What Qwen4-Max does
Qwen4-Max, often searched as Qwen Max 4, is Alibaba's new flagship Qwen model, announced by the Qwen team on 16 November 2026. It reads text, images and video, has a 1M-token context window and is built for long agent tasks, coding and office work. Alibaba says it has 2.4 trillion backbone parameters and uses fewer tokens per task than Qwen3.8-Max. Developers reach it through the Model Studio API or a Token Plan subscription, and Alibaba says it is also in the Qwen app and Qwen Studio.
Best for
- Developers who want a 1M-token model at a low per-token price
- Coding agents such as Claude Code, Cursor or Qwen Code on a flat monthly plan
- Long multi-step agent work that produces documents, spreadsheets and slides
- Teams already building on Alibaba Cloud Model Studio
Key features
1M-token context
The API model qwen4-max accepts up to about 990K input tokens and writes up to 131K tokens, or 262K in thinking mode.
Text, image and video input
It takes text, images and video and answers in text, with function calling, web search and structured JSON output.
Adjustable thinking
Thinking is on by default; you can set the effort to low, high or xhigh to trade speed and cost for depth.
Works with OpenAI and Anthropic SDKs
Model Studio serves it through OpenAI Chat Completions and Responses, Anthropic Messages and Alibaba's DashScope API.
Fewer tokens per task
On Alibaba's Work360 test it scored 52.69 against 47.74 for Qwen3.8-Max, and Alibaba reports about 24% to 35% fewer tokens used at xhigh and high effort.
Qwen4-Max pricing
In Singapore, the standard API costs $1.50 per million input tokens and $4.50 per million output tokens. These international rates apply to requests with up to 1M input tokens, with thinking on or off. In Beijing, Hong Kong, Frankfurt, Virginia and Tokyo, the standard API costs $1.413 per million input tokens and $4.24 per million output tokens. The same rates apply with thinking on or off. Qwen4-Max has no free plan you can keep using each month. New Model Studio users in the international (Singapore) region get a one-time 1 million free tokens, valid for 90 days. Token Plan Personal is a subscription for interactive use inside coding and agent tools only, not for scripts, app backends or batch jobs; it is sold in the Singapore region and cannot currently be cancelled once bought. Prices checked on the official site on Nov 17, 2026.
| Singapore | Input: $1.50 / 1M tokens Output: $4.50 / 1M tokens | These international rates apply to requests with up to 1M input tokens, with thinking on or off. |
|---|---|---|
| Beijing | Input: $1.413 / 1M tokens Output: $4.24 / 1M tokens | The same rates apply with thinking on or off. |
| Hong Kong | Input: $1.413 / 1M tokens Output: $4.24 / 1M tokens | The same rates apply with thinking on or off. |
| Frankfurt | Input: $1.413 / 1M tokens Output: $4.24 / 1M tokens | The same rates apply with thinking on or off. |
| Virginia | Input: $1.413 / 1M tokens Output: $4.24 / 1M tokens | The same rates apply with thinking on or off. |
| Tokyo | Input: $1.413 / 1M tokens Output: $4.24 / 1M tokens | The same rates apply with thinking on or off. |
| Token Plan Lite | $6/month | 20,000 credits a month; $3/month for new users until 30 Nov 2026 |
|---|---|---|
| Token Plan Essential | $10/month | 45,000 credits a month; $5/month for new users until 30 Nov 2026 |
| Token Plan Standard | $18/month | 80,000 credits a month |
| Token Plan Pro | $43/month | 200,000 credits a month |
Source: Qwen4-Max pricing page
Pros and cons
What it does well
- In Singapore, requests with up to 1M input tokens cost $1.50 per million input tokens and $4.50 per million output tokens.
- Large 1M-token context with text, image and video input.
- Drop-in for apps written against the OpenAI or Anthropic APIs.
Know before you sign up
- The preview model qwen4-max-preview stops working after 25 November 2026 (UTC+8); code that calls it must switch to qwen4-max.
- Token Plan cannot be used for automation scripts or app backends, and the Personal edition cannot be cancelled after purchase.
- Alibaba has not released Qwen4-Max weights; it says smaller open-weight Qwen4 models will follow.
Alternatives to Qwen4-Max
Gemini 4 CarbonFor long coding tasks or questions about a video, call Gemini 4 Carbon through the paid Gemini API. The Gemini app requires Google AI Ultra for this model.Claude Fable 5.2For coding agents and document analysis, call Claude Fable 5.2 through the Claude API. Input and output tokens are billed separately.
Questions about Qwen4-Max
What is Qwen4-Max?
Qwen4-Max (Qwen Max 4) is an AI tool from qwen.ai, listed in our AI assistants category. Alibaba's flagship Qwen model with a 1M-token context, for agents, coding and office work.
Is Qwen4-Max free?
New Model Studio users in the international (Singapore) region get a one-time 1 million free tokens, valid for 90 days.
How much does Qwen4-Max cost?
In Singapore, the standard API costs $1.50 per million input tokens and $4.50 per million output tokens. These international rates apply to requests with up to 1M input tokens, with thinking on or off. In Beijing, Hong Kong, Frankfurt, Virginia and Tokyo, the standard API costs $1.413 per million input tokens and $4.24 per million output tokens. The same rates apply with thinking on or off.
Where can I use Qwen4-Max?
On Web and API.
What are some alternatives to Qwen4-Max?
For long coding tasks or questions about a video, call Gemini 4 Carbon through the paid Gemini API. The Gemini app requires Google AI Ultra for this model. For coding agents and document analysis, call Claude Fable 5.2 through the Claude API. Input and output tokens are billed separately.