Available on request
Qwen API for teams
that need a specific Qwen.
Alibaba’s Qwen family covers general chat, coding and vision in one set of open-weight models. Tell us which Qwen you need and your expected volume, and we’ll talk you through adding it to the Xinference Model API.
Choose a model
Which Qwen fits the job
“Qwen” is a family, not one model. Most teams start by picking the one that matches the workload.
| If you need | Look at | Size |
|---|---|---|
| A general model that also reads images and video | Qwen3.8 | 27B, dense |
| A coding model for agents and repository-scale work | Qwen3-Coder | 480B total / 35B active, or 30B total / 3B active |
| Document, chart and screenshot understanding | Qwen3-VL | 235B total / 22B active, or 30B total / 3B active |
Sizes from the Qwen team’s model cards on Hugging Face (Qwen3.8-27B, Qwen3-Coder, Qwen3-VL), checked 23 October 2026. “Active” is the share of a Mixture-of-Experts model that runs for each token. Qwen and the Qwen logo are trademarks of Alibaba Cloud, shown only to identify the model. Their use does not imply endorsement.
Before you commit
Licence, context and thinking mode
Apache 2.0
The Qwen3.8, Qwen3-Coder and Qwen3-VL weights above are published under the Apache 2.0 licence, which allows commercial use. Check the model card of the exact version you pick.
262,144-token context
All three are listed with a 262,144-token context window (model maker’s figure). The limit we serve is agreed when we add the model.
Thinking you can switch
Qwen3.8’s makers describe flexible thinking control, so it can reason step by step when a task needs it and answer directly when it doesn’t.
How adding Qwen works
From request to first call
1. Tell us the model and volume
Which Qwen, which size, and roughly how many tokens a month. Volume decides how much capacity we set up.
2. We add it to the Model API
You call it through the same OpenAI-compatible API as Kimi K3 and DeepSeek V4.1 Flash.
3. Change one model name
Code already written for the OpenAI SDK keeps working. You point it at a new model name. What an OpenAI-compatible API covers.
Qwen API questions
Can I call a Qwen API on Xinference today?
Not as a self-serve model. The Model API’s standard models are Kimi K3 and DeepSeek V4.1 Flash, with public per-token prices. Qwen is available on request: we add it for teams with enough volume, so talk to us about the version and usage you have in mind.
Is Qwen free for commercial use?
The Qwen3.8, Qwen3-Coder and Qwen3-VL models listed on this page are released under Apache 2.0, which permits commercial use. Licences can differ between Qwen versions, so check the model card for the exact one you plan to run.
Which Qwen model is best for coding?
Qwen3-Coder is the Qwen line built for code. The 480B version with 35B active parameters targets agentic and repository-scale tasks; the 30B version with 3B active parameters is the lighter option when cost and speed matter more.
Can Qwen read images and PDFs?
Qwen3.8 is a native vision-language model that understands images and video, and Qwen3-VL is the family’s dedicated vision line. For scanned documents, charts and screenshots, Qwen3-VL is the usual starting point.
Qwen or DeepSeek: which should I use?
DeepSeek V4.1 Flash is a standard Model API model with public per-token pricing. Choose Qwen when you need a specific Qwen strength, such as the vision line or a smaller dense model you can size precisely.
Do you train on my data?
No. We don’t use your prompts or outputs to train or fine-tune models unless you agree in writing. See the Terms of Service.
Last updated