Neural network API
Qwen API: Alibaba Cloud or an open model on your server
The official Qwen API comes from Alibaba Cloud Model Studio: you create the key in the console, it works only in the region where you created it, and as of 16 October 2026 Model Studio has no regions in Russia. The second path is open Qwen 3 weights under the Apache 2.0 license, which you can run on your own server and connect through the same OpenAI format.
Access from Russia:The key is tied to a cloud region, and there are no regions in Russia(checked 2026-10-16, source)
In short: two paths to Qwen
The cloud fits if your requests contain no personal data and you need a quick start. Alibaba Cloud charges by actual use by default (pay-as-you-go), and in the Singapore region it gives a free quota for some models.
Open weights fit if the data is sensitive or the volume is large. On 29 April 2025 the Qwen 3 family came out: six dense models from 0.6 to 32 billion parameters and two mixture-of-experts (MoE) models, Qwen3–30B-A3B and Qwen3–235B-A22B. All eight are published under Apache 2.0.
The code is the same for both paths. Both Alibaba Cloud and a server with an open model accept requests in the OpenAI format, so moving between them does not require rewriting the app.
Alibaba Cloud Model Studio: key and regions
The key is created by the Alibaba Cloud account owner or a user with admin rights or access to the API-Key page. In the Model Studio console, choose a region in the top right corner, press Create API Key, and set the workspace and permissions. The full key is shown once, so save it right away.
The key is tied to the region where it was created. As of 16 October 2026, OpenAI-format endpoints exist in Singapore, the US (Virginia), China (Beijing), China (Hong Kong) and Japan (Tokyo). A key from one region is rejected by another region with an HTTP 401 error.
The region sets the access point and where data is stored, and the deployment scope sets where model inference runs. Static data always stays in the chosen region. The Model Studio region list includes Singapore, the US, Germany, China and Japan, and Russia is not on it.
Payment comes from the Alibaba Cloud account. As of 16 October 2026, the Alibaba Cloud help names Visa, Mastercard, AMEX and JCB bank cards with international payments turned on, and PayPal registered outside mainland China. One card can be linked to only one account.
Models and pricing
In the Model Studio model list (updated 11 October 2026), the main Qwen text models are qwen3.8-max, qwen3.7-plus and qwen3.8-flash. The same list has models for images and video, speech and voice, the text-embedding-v4 embeddings and the qwen3-rerank ranking model.
The price is given per million tokens, for most text models separately for input and output. For some models the price depends on a tier: the total number of input tokens in one request sets it, and the whole request is billed at that tier. Modes with and without reasoning often cost different amounts.
For models with a batch mode, batch calls cost 50% of the normal price. Creating an explicit cache costs 125% of the input token price, a cache hit 10%, and batch and cache discounts do not add up. Prices differ by region and even by deployment scope within one region.
A free quota exists only in Singapore: for example, qwen-max gets 1 million tokens for 90 days from the activation of Model Studio, the model release or approval of the request. Per-token figures by model are in the API price comparison, and the cloud API calculator shows the budget for your volume.
Limits apply to the whole Alibaba Cloud account: requests per minute (RPM) and tokens per minute (TPM) for each model separately, from all keys and workspaces together. When you exceed them you get HTTP 429, a balance top-up does not raise the limits, and for an increase you contact an Alibaba Cloud manager.
OpenAI compatibility
Model Studio accepts requests through an OpenAI-compatible interface. To move your code, change three things: the key, the address and the model name. As of 16 October 2026, the address contains the workspace ID and the region code, for example ap-southeast-1 for Singapore.
Your own server gives the same interface. vLLM runs OpenAI-compatible APIs: Chat, Completions, Responses, speech recognition and others. So code written for Alibaba Cloud moves to Qwen in your own rack by changing the address and the key.
It also works the other way: if the cloud does not fit on price or data requirements, you do not need to change the app. Change the address in the settings and check the model’s answers on your requests.
Open weights: Qwen on your own server
Apache 2.0 gives a perpetual, worldwide, free license to copy, create derivative works and distribute. Qwen 3 models range in size from 0.6 to 235 billion parameters, so you choose the version for the task and the hardware.
On your own server, requests are processed inside your network. This links directly to 152-FZ, Russia’s personal data law: since 1 March 2023 an operator must notify Roskomnadzor, the Russian data protection regulator, before a cross-border transfer of personal data, and the data of Russian citizens is collected and stored in databases in Russia.
Choose the Qwen version by testing it on your real requests: quality on your documents matters more than general rankings. Overview of open models and hardware requirements is in the section on sovereign AI.
If you have no server of your own, our service gives a dedicated API on Qwen, DeepSeek and other open-weight models. The format is OpenAI-compatible, you get access to the infrastructure and see the logs, and your requests are not used to train models.
Questions and answers
How do I get a Qwen API key?
Can I use Qwen in a commercial project?
Sources and dates
- Alibaba Cloud Model Studio, OpenAI compatibility (help.aliyun.com), checked 16 October 2026help.aliyun.com
- Alibaba Cloud Model Studio, OpenAI compatibility (updated 13 October 2026), checked 16 October 2026alibabacloud.com
- Alibaba Cloud Model Studio, Models (updated 11 October 2026), checked 16 October 2026alibabacloud.com
- Alibaba Cloud Model Studio, Model pricing (updated 15 October 2026), checked 16 October 2026alibabacloud.com
- Alibaba Cloud Model Studio, Rate limits (updated 16 October 2026), checked 16 October 2026alibabacloud.com
- Alibaba Cloud Model Studio, Get API key (updated 11 October 2026), checked 16 October 2026alibabacloud.com
- Alibaba Cloud Model Studio, Regions (updated 13 October 2026), checked 16 October 2026alibabacloud.com
- Alibaba Cloud, Payment FAQ (updated 7 September 2026), checked 16 October 2026alibabacloud.com
- Simon Willison, Qwen 3 (29 April 2025), checked 16 October 2026simonwillison.net
- Qwen, Qwen3 blog (29 April 2025), checked 16 October 2026qwenlm.github.io
- Apache License 2.0, section 2, checked 16 October 2026apache.org
- vLLM Docs, OpenAI-compatible server, checked 16 October 2026docs.vllm.ai
- Brace Law Firm, cross-border transfer of personal data, checked 16 October 2026brace-lf.com