Names, versions and owners change fast. Checked October 2026.
In plain words
Qwen is the name of a family of AI models, not a single app. It is made by Alibaba, the Chinese e-commerce and cloud company, through its cloud division. Qwen is short for its Chinese name, Tongyi Qianwen.
Each Qwen model is a large language model: a program that learned from huge amounts of text and can chat, write, translate, code and answer questions. Some versions also understand pictures and video, which is called multimodal AI.
What makes Qwen well known outside China is that Alibaba publishes many of its models with open weights, often under the permissive Apache 2.0 licence. Developers anywhere can download them, run them on their own computers and adapt them. Its largest, most capable versions are usually kept closed and offered through Alibaba's cloud instead.
Everyday example
A bakery that shares recipes
Picture a big bakery chain. It sells its fanciest cakes only in its own shops. But it also publishes the recipes for many of its everyday breads, in small, medium and large batch sizes, so home bakers and small cafés can bake them themselves and add their own twist. Qwen works like that: the biggest models stay in Alibaba's cloud, many others are free to download.
Where the example breaks: A recipe lists every ingredient. Open weights give you the finished model's numbers, not the full training data or every step used to make it.
Latestas of October 2026
- Newest family
- Qwen3.8, from August 2026, with open-weight sizes under Apache 2.0 and a larger Qwen3.8-Max
- Next generation
- Alibaba said at its 2026 Apsara conference that Qwen 4 is in training
- Earlier closed flagship
- Qwen3.7-Max, May 2026, described by Alibaba as proprietary
These details change often. The rest of this page explains what it is, which stays true.
How it actually works
Qwen is less one model than a whole range, released in numbered generations.
Many sizes in each generation
Each generation comes in sizes from small models that fit on a laptop to very large ones that need a server full of GPUs. Size is measured in parameters, the numbers the model learned.
Open and closed side by side
Smaller and mid-sized versions are usually released as open weights on sites such as Hugging Face. The top "Max" versions are proprietary and reached through Alibaba Cloud's API.
Thinking when needed
Recent Qwen models can work through hard problems step by step before answering, like a reasoning model, and this thinking mode can be switched off for quick replies.
- Made by
- Alibaba, through Alibaba Cloud
- Full name
- Tongyi Qianwen
- First open release
- Qwen-7B and Qwen-7B-Chat, August 2023
- Licence
- Many models under Apache 2.0; the largest Max models are closed
Other well-known open-weight families include Llama, DeepSeek and Mistral.
Questions people ask
Is Qwen open source?
Many Qwen models have open weights under the Apache 2.0 licence, so you can download, use and change them, including commercially. The training data is not published, and the largest Max models are closed, so "open weights" is the more exact term.
Is Qwen the same as DeepSeek?
No. Both are Chinese model families that publish open weights, but they come from different companies. Qwen is made by Alibaba, a large technology group. DeepSeek is made by a separate company in Hangzhou backed by an investment fund.
Can I run Qwen on my own computer?
The smaller versions, yes. With free tools you can download a small Qwen model and chat with it offline on a modern laptop, especially after quantization shrinks it. The largest versions need powerful GPUs in a data centre.
Who makes Qwen?
Alibaba, the Chinese technology group best known for online shopping and cloud computing. Its cloud division develops the Qwen models, releases many of them openly, and sells access to the largest ones through Alibaba Cloud.