The Qwen3 Max family
Against 3.7, Qwen3.8 Max gained three things, image and video input, structured outputs and double the output ceiling, and lost one that rarely gets written down: Batch Inference is not supported on it. If your workload is batched, the upgrade is a step backwards.
- image and video input on 3.8
- the output ceiling doubled
- Batch Inference was lost
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. What changed
Take 3.7 or 3.8
If your input is text only and you send the work in batches, stay on 3.7 Max. If you feed images or video, need structured outputs, or want an answer longer than 65K tokens, take 3.8 Max. This is the only family in this reference where the answer to which version is not a single name.
The figures come from each model own page. Both carry a 1,000,000 token context window and both cap input at 991,808 tokens, so that column separates nothing. Maximum output is 65,536 tokens on 3.7 and 131,072 on 3.8, exactly double. The chain-of-thought ceiling is 262,144 tokens on both.
The capability table is where the difference actually lives. 3.7 Max accepts text input only and its page describes itself as a pure text interface; 3.8 Max accepts image, text and video. Structured outputs are unsupported on 3.7 and supported on 3.8. And Batch Inference runs the other way: supported on 3.7, unsupported on 3.8.
The capability the upgrade took back
That last row is worth stopping on. The 3.7 Max page does not merely mark Batch Inference as supported, it carries separate prices for it: batch file input at $0.825 and batch output at $2.475, half the standard rate. On the 3.8 Max page, Batch Inference reads unsupported in all six regions and there is no batch price row at all.
For anyone processing a few thousand documents overnight in batches, that means upgrading to 3.8 doubles the unit cost without a single figure changing in the headline price column. This kind of regression does not show up in comparison lists, because those lists read the price column and not the capability column.
The price depends on region, and a temporary discount sits in the middle of it
The 3.8 Max page prices six regions separately. Beijing, Frankfurt, Virginia, Tokyo and Hong Kong all sit at $1.65 input and $4.951 output. Singapore, which is the international region, sits at $2 and $6. The figure we carry in our tables is that international row.
On the central pricing page, 3.7 Max shows an international list price of $2.5 and $7.5 with a limited-time fifty percent discount noted beside it. So buying today may cost half of that, and budgeting six months out may not. We always take the list price, because a limited-time discount is not a number you can write a contract against.
How this model was found, and why that is the content
Until a few days ago the price and context columns for this model read unverified in our table, and we wrote the reason plainly: the English Model Studio catalogue and the central pricing page, both last updated 15 July 2026, name no qwen3.8-max, while the Arena leaderboard row carries exactly that name. Assuming 3.8 was simply 3.7 would have been easy, but it was a guess, and now that the two pages sit side by side we know it was a wrong one.
What turned up next: the model has its own dedicated page, last updated 3 August 2026. So Alibaba index pages had fallen behind its models rather than the model not existing. There is one wrinkle that makes it hard to find: the URL slug is written with a hyphen while the model id uses a dot.
The practical lesson for anyone chasing the specs of a Chinese model: the vendor list is not the final source. Try the model own page address, even when it is absent from the list.
Specifications
Every number here comes from the vendor's own page, with that page linked beside it.
Release timeline
Each version with its own release date, and what changed against the one before it.
- Qwen3.8 Max current
Using it from Iran
This column carries a date and says how it was checked, because most listicles guess it. Where we have only read a vendor policy page, the note below says exactly that.
- Reachable
- not checked
- Payment
- not checked
- Free tier
- no
How we checked: Alibaba Cloud publishes no blocked-countries list for this service anywhere we could find, and we have not tested from an Iranian connection. So the access column stays unchecked.
What it is good at
- Multimodal input on 3.8 Max: image, text and video, where 3.7 is text only
- A 131,072 token output ceiling on 3.8, double that of 3.7
- Structured outputs are supported on 3.8
- The same model charges $1.65 input across five global regions, below the international row
- The model own page breaks out max input and thinking-mode limits, which no other vendor in this group publishes
Where it falls short
- Batch Inference is unsupported on 3.8 Max while it is supported on 3.7, and the 3.7 batch price row was half the standard rate. For a batched workload the upgrade is a regression.
- Fine-tuning is unsupported on both members of this family.
- The Alibaba catalogue and pricing pages still do not list qwen3.8-max, so a buyer reading only those two pages concludes the model does not exist.
- The 3.7 Max international price carries a limited-time fifty percent discount, so today real figure differs from the list figure and no end date is published.
- We have not measured access from Iran, and Alibaba publishes no list we could find.
Our take
If you send text only and work in batches, stay on 3.7 Max and skip the upgrade, because the one thing you would lose is the thing that was halving your bill. For anything else take 3.8 Max, and read its model id off the model own page rather than the catalogue.
Questions people actually ask
Is Qwen3.8 Max the same model as Qwen3.7 Max
No. Alibaba own pages differ in five places: input modality (text only against image, text and video), structured outputs, maximum output (65,536 against 131,072), Batch Inference, and the international price row. The context window and max input length are identical.
Why is Qwen3.8 Max missing from the Alibaba catalogue
The catalogue and the central pricing page were both last updated on 15 July 2026, while the model own page was updated on 3 August. So the lists have fallen behind the model. The model page follows the same URL pattern as the others; its slug is just written with a hyphen where the model id uses a dot.