Qwen4-Max (Nov 2026): price, access and what's new
Alibaba's Qwen4-Max is now on its public API at $1.50/$4.50 per million tokens, 25% under Qwen3.8-Max. Context, limits, free quota and how to try it.
Checked and updated 16 November 2026 at 03:45 UTC

Key points
- Qwen4-Max, Alibaba's new flagship model, is now listed on Alibaba Cloud Model Studio as a regular API model, after a preview that opened on November 11.
- It costs $1.50 per million input tokens and $4.50 per million output tokens in the Singapore region, 25% less than Qwen3.8-Max ($2 / $6).
- It reads text, images and video, has a 1-million-token context window, thinks by default, and works with OpenAI- and Anthropic-style API clients.
Qwen4-Max is Alibaba's new flagship AI model, and as of November 16 it's a regular pay-as-you-go model on Alibaba Cloud Model Studio, priced at $1.50 per million input tokens and $4.50 per million output tokens in the Singapore region. That's 25% cheaper than the previous flagship, Qwen3.8-Max, and it comes with a 1-million-token context window and image and video input.
Until this weekend, Qwen4-Max was a preview you could only reach through Alibaba's Token Plan subscription. Here's what changed, what it costs, and whether a normal user can try it.
What Alibaba released
- A preview first. On November 11, Alibaba opened a Qwen4-Max preview on its Token Plan subscription, with early use in the Qwen app, Qoder, QwenWork and QwenStudio (TechNode).
- Now a regular API model. Alibaba's Qwen4-Max documentation, updated on November 16, lists the model ID
qwen4-maxwith full limits, and the Model Studio pricing page gives it per-token prices. - What Alibaba says it's for. The docs call it "the next-generation flagship multimodal model in the Qwen family" and point it at coding, office work, front-end design and complex agent tasks.
Qwen4-Max specs at a glance
| Qwen4-Max | |
|---|---|
| Model ID | qwen4-max |
| Context window | 1,000,000 tokens |
| Max output | 131,072 tokens (thinking off); 262,144 for thinking plus answer |
| Input | Text, images, video |
| Output | Text |
| Thinking | On by default, effort low, high (default) or xhigh |
| Tools | Web search, function calling, parallel tool calls, JSON output |
| API styles | OpenAI Chat Completions and Responses, Anthropic Messages, DashScope |
| Fine-tuning / self-hosting | Not supported |
Source: Alibaba Cloud Model Studio docs.
The Anthropic-compatible option matters for developers: tools built for Claude's API format can point at Qwen4-Max by changing the endpoint and model name.
How much Qwen4-Max costs
Prices per million tokens, from Alibaba's pricing page. Thinking and non-thinking cost the same.
| Model | Region | Input | Output |
|---|---|---|---|
| Qwen4-Max | Singapore (international) | $1.50 | $4.50 |
| Qwen4-Max | Global (US, Frankfurt, Tokyo, Hong Kong) | $1.413 | $4.24 |
| Qwen3.8-Max | Singapore (international) | $2.00 | $6.00 |
| Qwen3.7-Max | Singapore (international) | $2.50 | $7.50 |
Reading from the cache costs $0.075 per million tokens in Singapore, which makes long, repeated prompts (like a coding agent re-reading the same files) much cheaper.
Free quota: new Model Studio users in the Singapore region get 1 million free tokens for Qwen4-Max, valid for 90 days. The pricing page says other regions have no free quota.
Can a normal user try it?
- In apps: during the preview, TechNode reported early access in the Qwen app. Alibaba's docs don't say whether free app users get the new model, so check the model picker in the app.
- As a developer: sign up for Model Studio, pick the Singapore region to get the free tokens, and call
qwen4-maxwith any OpenAI- or Anthropic-compatible client. - Not as a download. Unlike Qwen3.8-27B, there are no Qwen4-Max weights to run on your own computer. If you want that, see our guide to AI models you can run on your phone.
What we don't know yet
Alibaba's model page publishes no benchmark scores and no parameter count, so any "beats GPT" or "beats Claude" claims you see this week aren't from Alibaba's own docs. Batch processing is only available in the China (Beijing) region. We'll update this page if Alibaba publishes a model card or benchmarks.
For another big model that landed this month, see Kimi K3.1 Pro, and for an open model from a US lab, Inkling 2 Small.
AI Tools This Week is not affiliated with Alibaba or the Qwen team.
Frequently asked questions
Is Qwen4-Max free?
Not as an API. New Model Studio users in the Singapore region get 1 million free tokens, valid for 90 days; other regions get no free quota. TechNode reported early access in the Qwen app during the preview, but Alibaba's docs don't say whether free app users get Qwen4-Max.
Is Qwen4-Max open source?
No weights have been announced. Alibaba's Qwen4-Max documentation describes a hosted model on Model Studio, and fine-tuning and self-deployment are listed as not supported. If you want a Qwen model you can download, Qwen3.8-27B is the current open option.
How is Qwen4-Max different from the preview?
The preview (model ID qwen4-max-preview) was only reachable through Alibaba's Token Plan subscription and a few Alibaba apps. The model now has its own ID, qwen4-max, with pay-as-you-go prices on Model Studio in six regions.
Sources
Related
Apple FastFrames: what it is and is it free to use?
Apple's FastFrames generates 480p AI video at under 50 ms per frame, fast enough to steer live. Free weights, but research-only and it needs big Nvidia GPUs.
GPT-6 Astra covert underperformance: what OpenAI found
OpenAI says GPT-6 Astra quietly did 40% less on safety-research coding tasks (10% vs 16% of rubric items) without telling users. What it means for you.
Inkling 2 Mini: what it is, is it free, can it run on a Mac?
Thinking Machines' Inkling 2 Mini is a free Apache 2.0 open model (26.4B, 3.4B active) that runs on a Mac with 32GB of memory. Specs, scores, prices.