Qwen3.7 Max Costs Half as Much on Novita as on Alibaba Cloud

Qwen3.7 Max costs half as much on Novita AI as it does on Alibaba Cloud, the company that built it. In our tracked prices, Novita charges $1.25 per million input tokens and $3.75 per million output tokens. Alibaba's own platform charges $2.50 and $7.50. Same model, same token, double the bill.

If you pay Alibaba's list price for Qwen3.7 Max today, you are probably overpaying.

The numbers

Here is one realistic month for a product with an AI feature: 10 million input tokens and 2 million output tokens.

Host	Input / 1M	Output / 1M	Cost for that month
Novita AI	$1.25	$3.75	$20.00
Together AI	$1.50	$4.50	$24.00
Alibaba Cloud	$2.50	$7.50	$40.00
DeepInfra	$2.50	$7.50	$40.00

(Our calculation from the tracked list prices. Cache discounts are not included, and your own input-to-output mix will move the totals.)

The cheapest host saves $20 a month on this workload, and the saving grows linearly with traffic. At 10 times the volume that is $200 a month, for changing a base URL and an API key.

Why it is not just "pick the cheapest"

I would not wire up Novita blindly, and neither should you. Three things to check first:

• We track list prices, not quality. A host can serve a different quantization, a different context limit or lower rate limits at a lower price. Our tables do not show that. Test your own prompts on the cheap host before you switch.
• The cheaper price may not last. Together AI is a good example. Our change log shows it moved Qwen3.7 Max from $1.50 to $2.50 on October 11, then back to $1.50 on October 15. Prices on this model are moving, so check the date on any number you rely on.
• The maker's price is not always the floor. Alibaba is the model's creator, yet it is the most expensive row in the table. Resellers and hosts compete on price. Model creators often do not.

What I would do

If you use Qwen3.7 Max in production, compare at least Novita and Together against your current host with your own traffic. If quality holds up, the saving can outweigh the migration effort quickly at high volume; at low volume it may not.

If you are only experimenting, it does not matter. The difference on a few hundred thousand tokens is cents.

If you are not locked into Qwen at all, run the numbers for other models too. Our model page for Qwen3.7 Max shows the live prices for every host we track, and the calculator there turns your own token volume into a monthly bill. The October 2026 price report shows which other models and hosts moved this month.

The prices behind this take

The Novita and Alibaba prices were confirmed on their official pricing pages on October 17, 2026; the Together and DeepInfra prices come from our daily price tracking, which reads the open LiteLLM and models.dev datasets and links to each host's official pricing page. The official pages for Novita AI, Together AI, Alibaba Cloud and DeepInfra are where you should confirm the number before you commit. We say "tracked" on purpose: these are prices we have seen consistently in the data, not a promise from the provider.

The honest summary: one model, a 2x price gap between hosts, and a maker that is not the cheapest. That is why we compare hosts instead of just models.