Alibaba’s new AI model, Qwen4-Max, looks like a bargain. Released on 16 November 2026, it charges US$4.50 for every million tokens it writes. OpenAI charges US$10 for GPT-6.2 Sol, more than twice as much.
In independent testing, though, the bargain disappeared.
Artificial Analysis, a company that runs its own tests of AI models, put both through the same tasks. Qwen4-Max cost more than three times as much per task as GPT-6.2 Sol. It also took more than three times as long, and it scored lower.
Why a cheaper price can mean a bigger bill
AI companies charge by the token: the small chunks of text a model reads and writes. Your bill depends on two things: the price of each token, and how many tokens the model uses to finish the job.
Qwen4-Max used far more tokens in the tests. It read more than four times as many input tokens as GPT-6.2 Sol and wrote about twice as many output tokens. Its cost per task came out higher despite its lower prices.
Price is not the whole picture
Of the three models compared, Anthropic’s Claude Opus 5.5 was the most expensive, at about three times Qwen4-Max’s cost per task. It also scored highest. Qwen4-Max scored lowest of the three and was the slowest.
These results come from one company’s tests. Its scoring combines four areas: knowledge work, coding, science and general questions. Your own work may give different results.
Separately, Alibaba says Qwen4-Max cuts the average cost per task by around 60% compared with its previous model. That compares Alibaba’s models with each other, not with its rivals.
What this means for you
These prices apply to businesses and developers who pay for AI by the token, not to people choosing a monthly chatbot subscription.
If you are comparing AI services, the price list is only the start. What counts is the cost of getting a job done, how long it takes and how good the result is. The surest way to find out is to try the models on your own work.
The numbers
| Qwen4-Max | GPT-6.2 Sol | Claude Opus 5.5 | |
| Output price, per million tokens | US$4.50 | US$10.00 | US$20.00 |
| Average cost per task in testing | US$3.27 | US$1.01 | US$10.04 |
| Average time per task | 31.4 minutes | 8.7 minutes | 25.2 minutes |
| Test score (higher is better) | 43 | 57 | 59 |
Prices are standard list rates (Alibaba’s international service; OpenAI’s standard rate for shorter prompts). Test figures are from Artificial Analysis’s Intelligence Index on 19 November 2026. GPT-6.2 Sol and Claude Opus 5.5 were tested at Max effort (Claude Opus 5.5 at Max, Default Fallback).
On paper, Qwen4-Max is the cheapest of the three. In testing, the cheapest job was done by GPT-6.2 Sol, which charges more than twice as much for the text it writes.
The real test for Alibaba
In a commentary on 17 November, the Chinese business paper National Business Daily said the middle of the market is the hardest place to be. Below Qwen4-Max, open models are fighting a price war; above it, closed models set the standard for capability. Whether Alibaba can win on both fronts, the paper said, will show in how many business customers actually switch over the following two quarters, not in launch-day test scores.
Related article: Chinese student’s dorm AI animated-drama business tops 3 million yuan






