What Alibaba sells in AI
Alibaba sells the Qwen models through a service called Model Studio, and the newest model on that line is qwen3.8-max: a one million token window, up to 131,072 tokens of output, at $2 in and $6 out in the international region. The catalogue is the deepest in this group and it has also fallen behind its own models.
- one million token window
- price depends on region
- separate embedding and reranking models
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. What changed
Model families
The Qwen Max line, which is where the list fell behind the model.
Versions
| Versions | Released | Status | Model families |
|---|---|---|---|
| Qwen3.8 Max | current | Qwen3 Max |
The window is one million tokens, but your input is not
The qwen3.7-max page publishes four separate numbers where the rest of this group publishes one: a context window of 1,000,000 tokens, a maximum input of 991,808, a maximum output of 65,536, and in thinking mode an input of 983,616 with a chain-of-thought ceiling of 262,144.
The split has practical value. Every marketing line about a million token window looks the same, yet what you can actually send is smaller, and smaller again once thinking is on. Alibaba is the only maker here that tells you, which is why we keep the maximum input figure in the text.
The price depends on where you call from
The pricing page gives $2.5 in and $7.5 out for the international region, with no tiering up to a million tokens. Two other models on the same line do tier: qwen3.7-plus runs $0.4 and $1.6 below 256K tokens and $1.2 and $4.8 above it, and qwen3.6-flash moves the same way, from $0.25 and $1.5 to $1 and $4. The same prompt, once it grows, is billed at three times the rate.
The model page itself shows price tables per region, and the China (Beijing) row is lower. That row does not state its currency on the page, so we do not print it and we treat the international figure as ours. It is worth saying, because a round-up that quotes one single price for Qwen has not told you which region it read.
A catalogue that reaches past the chat model
Beside the three text models, Model Studio lists embedding models (text-embedding-v4 and tongyi-embedding-vision-plus) and a reranking model (qwen3-rerank), plus the multimodal qwen3.5-omni-plus. None of the other three Chinese makers in this reference lists anything of the sort. If you are building semantic search or a retrieval pipeline, that difference matters more than a few leaderboard points.
A list that has fallen behind its own models
The Alibaba row on the Arena leaderboard is called qwen3.8-max, and the English Model Studio catalogue lists no model of that name. Both the catalogue and the central pricing page carry a last-updated date of 15 July 2026.
The model itself does exist. Its dedicated page sits on the same URL pattern as every other model, was updated on 3 August 2026, and gives everything: context window, maximum input and output, a capability table and prices for six regions. So the problem was never that the model does not exist. Alibaba index pages have fallen behind its models.
And the two are not the same model. qwen3.8-max takes image and video input where 3.7 describes itself as a pure text interface, its output ceiling is double, and Batch Inference, which works on 3.7, does not work on it. The full comparison sits on the family page.
What it is good at
- The deepest catalogue among the Chinese makers in this reference: text, multimodal, embedding and reranking models in one service
- The only maker that publishes maximum input and a chain-of-thought ceiling separately from the context window
- The API accepts the OpenAI format, the Anthropic format and its own DashScope format
- A one million token window on qwen3.8-max at $2 and $6 in the international region, plus image and video input
Where it falls short
- The official catalogue and the central pricing page still list no model called qwen3.8-max, which is exactly the leaderboard row and the name on the model own dedicated page. A buyer reading only those two pages concludes the model does not exist.
- Maximum output on qwen3.7-max is only 65,536 tokens, half the qwen3.8-max ceiling and the lowest in this group against 384K for DeepSeek.
- Batch Inference is unsupported on qwen3.8-max where it worked on qwen3.7-max, and the 3.7 batch price row was half the standard rate.
- Price depends on region, and the model page carries separate tables for Beijing, Frankfurt, Virginia, Tokyo and Hong Kong. There is no single number for the price of Qwen.
- Two other models on the line triple in price above 256K tokens, so the cost of a long prompt is not linear.
- It publishes no supported-countries list, so our Iran column stays unchecked.
Our take
If you are building a pipeline that needs embedding and reranking alongside a chat model, Alibaba is the only one-stop option in this group. If you are shopping for the model you saw on the leaderboard, read its id off the model own page rather than the catalogue, because the catalogue lags.
Questions people actually ask
What does qwen3.7-max cost
In the international region its list price is $2.5 per million input tokens and $7.5 per million output tokens, with no tiering up to a million tokens, and a limited-time fifty percent discount is noted beside it. The model page gives $1.65 and $4.951 for the other regions, so what you pay depends on the region you call.
Where is qwen3.8-max in the catalogue
It is not in the catalogue, but it does have its own dedicated page, updated on 3 August 2026. The catalogue and the pricing page were both last updated on 15 July, so they are older than the model. One wrinkle: the URL slug is written with a hyphen while the model id uses a dot.