Qwen4-Max Is Alibaba’s New Flagship AI, and It Makes GPT and Claude Look Expensive
Alibaba Qwen4-Max costs $1.50 per million input and $4.50 per million output tokens, far below Claude Opus 5.5 and GPT-6 Astra. Specs, prices and caveats.

Alibaba released Qwen4-Max on November 16, calling it the most capable Qwen model it has built. The bigger story for developers, though, is the price tag. At $1.50 per million input tokens and $4.50 per million output tokens, the new Chinese flagship costs a fraction of what Anthropic and OpenAI charge for their top models.
Quick answer: Qwen4-Max is Alibaba’s new flagship AI model, with 2.4 trillion parameters and a 1 million token context window. Its API costs $1.50 per million input tokens and $4.50 per million output tokens. Claude Opus 5.5 costs $4 and $20, and OpenAI’s GPT-6 Astra costs $10 and $50.

How big is the price gap?
The list prices, per million tokens:
| Model | Input | Output |
|---|---|---|
| Qwen4-Max (Alibaba) | $1.50 | $4.50 |
| Claude Opus 5.5 (Anthropic) | $4.00 | $20.00 |
| GPT-6 Astra (OpenAI) | $10.00 | $50.00 |
Output is where it really shows. Qwen4-Max charges less than a quarter of Claude Opus 5.5 and less than a tenth of GPT-6 Astra. To be fair to OpenAI, its mid-tier GPT-6.1 Sol runs $2 and $10, which narrows the gap a lot.
Prices also shift by region. Alibaba Cloud’s pricing page lists $1.413 input and $4.24 output for its US (Virginia), Frankfurt and Tokyo regions, and the Singapore region comes with 1 million free tokens for 90 days.
What Alibaba says the model can do
According to Alibaba’s announcement, Qwen4-Max is a mixture-of-experts model with 2.4 trillion total parameters, about 71 billion of which are active for any given token. It reads images natively and can take in up to 1 million tokens at once.
Alibaba says it ranked sixth on the Code Arena coding leaderboard. On Work360, Alibaba’s own test across 721 job fields, it says the average cost per task fell about 60% compared with Qwen3.8-Max, partly because the model uses 35% fewer tokens at high reasoning effort.
A caution here. Most of these numbers come from Alibaba itself, and Work360 is its own benchmark. Independent testing will matter more than launch charts.
What to know before you switch
- It is closed: Qwen4-Max does not ship with open weights. Alibaba says smaller open Qwen4 models are coming “soon,” with no date.
- Where to use it: QwenCloud through the API and Token Plan, plus QwenWork and QwenStudio.
- Data location: US businesses should check which region they call. Some companies have rules about sending data to providers based in China.
The price pressure lands at a busy moment. Anthropic is preparing its IPO, Nvidia reports earnings today, and Claude users hit outages just last week.
FAQs
How much does Qwen4-Max cost?
Alibaba lists $1.50 per million input tokens and $4.50 per million output tokens in its Singapore region, and slightly less in its US region.
Is Qwen4-Max open source?
No. Alibaba says smaller open-weight Qwen4 models will follow later.
Is Qwen4-Max better than Claude or GPT?
Alibaba claims strong coding and agent results, but most published scores come from Alibaba. Independent tests are still limited.
Featured image: Jerry Chen on Unsplash.
Disclosure: TheTechHacker has no paid relationship with Alibaba, Anthropic or OpenAI.
More in News

OpenAI Adds Tougher ChatGPT Penalties to Stop Child Exploitation, Releases Free Detection Tool
ChatGPT now warns, deletes the chat and pauses access for 24 hours when someone seeks child abuse material. OpenAI also released a free detection classifier.

Nvidia Earnings Today: Time, the $108 Billion Forecast and What to Watch After Jensen Huang’s Exit
Nvidia reports Q3 results Nov 17 at about 4:20 p.m. ET, its first since Jensen Huang resigned. The $108 billion forecast, call times and what to watch.

Papa Johns Is Using AI to Check Your Delivered Pizza, and Paying You in Pizza for Photos
Papa Johns asks customers to photograph delivered pizzas for AI quality checks and gives a free pizza in return. How it works, the rules and the AI details.
The week in tech, minus the noise.
New reviews, fixes and the stories that matter — once a week, straight to your inbox.
We only use your email to send the newsletter. Unsubscribe any time. See our privacy policy.

Comments