AI Development
Qwen4-Max in Claude Code: Alibaba’s model that costs less than a quarter of Opus, and how to set it up in 3 steps
Qwen4-Max costs $4.5 per million output tokens versus $20 for Opus 5.5, and you can use it in Claude Code in 3 steps: Token Plan, API key and one prompt.

Luca Di Domenico
· 4 min read
In short
Qwen4-Max, Alibaba’s new flagship model released on November 16, 2026, costs $4.5 per million output tokens versus $20 for Claude Opus 5.5. On the Artificial Analysis index it scores 43 against 59, but on Code Arena WebDev it is close to Opus: 1,750 against 1,816. You can use it in Claude Code in 3 steps: Token Plan on QwenCloud, an API key and a prompt that configures settings.json.
Qwen4-Max is Alibaba’s new flagship model, released on November 16, 2026, and you can use it inside Claude Code in 3 steps. Output costs $4.5 per million tokens, against $20 for Claude Opus 5.5: less than a quarter. On independent benchmarks it sits a step below Opus, but on everyday coding it gets very close.
What is Qwen4-Max?
Qwen4-Max is the most capable model in Alibaba’s Qwen family. It reads text, images and video, and has a 1 million token context window: it can work on a large project in one go, without splitting it into pieces.
According to Alibaba’s official announcement, it is a Mixture-of-Experts model with 2.4 trillion parameters, of which 71 billion are active per token. Fewer active parameters make each answer cheaper to compute, which is why the price stays low.
How much does Qwen4-Max cost compared to Claude Opus 5.5?
Here are the prices per million tokens, from the official price lists:
- Qwen4-Max: $1.5 input and $4.5 output, from the model page on QwenCloud;
- Claude Opus 5.5: $4 input and $20 output, from Anthropic’s pricing page.
On output, Qwen4-Max costs 22.5% of Opus 5.5, so less than a quarter. On input it costs a little over a third. Caching cuts the bill further: tokens already read cost $0.075 per million.
Is Qwen4-Max as good as Opus?
Not quite. On the Artificial Analysis Intelligence Index, an independent ranking, Qwen4-Max scores 43. Claude Opus 5.5 at maximum effort scores 59, and even at medium effort it reaches 48.
On everyday coding the gap narrows. In the Code Arena WebDev leaderboard reported by Alibaba, Qwen4-Max ranks sixth with 1,750 points, against 1,816 for Opus 5.5 (Max). Only models from Anthropic, Google and OpenAI are ahead of it.
So Qwen4-Max makes sense for daily work, when you want to spend less. For the hardest tasks, Opus stays ahead.
Which other Qwen 4 models are coming?
In September Alibaba announced three more models in the Qwen 4 family:
- Flash, the fastest;
- Plus, halfway between Flash and Max;
- Qwen 4 with 27 billion parameters, which you will be able to download and run on your own computer.
As of November 17, 2026, none of the three has been released. In the Qwen4-Max announcement, Alibaba says the smaller open-weight models in the series will be released soon.
How do you use Qwen4-Max inside Claude Code?
It takes 3 steps.
1. Activate the Token Plan on QwenCloud
Go to QwenCloud.com and activate the Token Plan, the subscription that includes Qwen4-Max. The base plan costs a few dollars a month: on the page I show, the Lite plan is $6 a month, down from $8, with 20,000 credits. It is plenty for experimenting with the model.
2. Create an API key
Still in the QwenCloud web app, open the API Keys section, create a key and copy it. The key works like a password: never share it.
3. Ask Claude Code to configure itself
Open Claude Code and send it this prompt, with your key in place of the placeholder:
Configure my settings.json to use Qwen's Token Plan with the qwen4-max model. Address: https://token-plan.maas.qwencloudapi.com/apps/anthropic
My key: <your sk-sp-... key>Claude Code adds three settings to ~/.claude/settings.json: ANTHROPIC_BASE_URL with Qwen’s address, ANTHROPIC_AUTH_TOKEN with your key and ANTHROPIC_MODEL set to qwen4-max.
Then close and reopen Claude Code and type /status. If the Model line shows qwen4-max, you are done.
How do you switch back to Anthropic’s models?
Ask Claude Code to remove the configuration, for example: "Remove the Qwen configuration from my settings.json". Claude Code deletes the lines it added, and when you reopen it you are back on Anthropic’s models.
Before connecting another provider’s model to your company’s code, ask yourself where your prompts and files end up. I covered this for Kimi K3.1 Pro in Claude Code, a very similar case.
Is it worth trying?
Qwen4-Max does not replace Opus on the hardest tasks. For everyday coding, though, it costs less than a quarter on output and comes close. For a few dollars a month you can try it on your own projects and decide where it is enough for you.
Sources
- Alibaba Unveils Qwen4-Max: New AI Teammate for Work, Creativity, and Discovery — Official announcement, November 16, 2026: 2.4 trillion parameter MoE with 71 billion active per token, 1 million token context, 6th on Code Arena (1,750 vs 1,816 for Opus 5.5 Max in the chart), available on QwenCloud via Token Plan and API, smaller open-weight models coming soon
- Qwen4-Max — QwenCloud — Model page read on November 17, 2026: $1.5 input and $4.5 output per million tokens, $0.075 for cached input, 1M context, text, image and video input
- Pricing — Claude Platform Docs — Official price list read on November 17, 2026: Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens
- Qwen4 Max — Artificial Analysis — Read on November 17, 2026: Qwen4 Max scores 43 on the Artificial Analysis Intelligence Index
- Claude Opus 5.5 — Artificial Analysis — Read on November 17, 2026: Claude Opus 5.5 scores 59 at max effort, 57 xhigh, 52 high, 48 medium, 35 low
Want to know what this means for your product?
If you're building software with AI and want a second opinion on how you're working, tell me about your project.
Tell me about your projectFollow the blog
One email for every new article
I only write when an article is out: title, summary and link. You can unsubscribe with one click from any email.
RSS feed
An RSS feed is the up-to-date list of articles: add it to a reader app like Feedly or Inoreader and new articles show up there, no email needed.