Qwen: a plain-English guide

What Qwen is and who makes it

Qwen is a family of AI models built by the Qwen team at Alibaba, the Chinese technology group behind Taobao, Alipay and Alibaba Cloud. In Chinese it is also known as Tongyi Qianwen. Qwen covers chat models, coding models, models that understand images and video, image generators and safety classifiers, and it powers Alibaba's own assistant, also called Qwen.

Qwen stands out for how much of it is published openly. Alibaba has released a very large number of open-weight models in many sizes, most under the permissive Apache 2.0 license, and Qwen has become one of the most widely used open model families among developers worldwide. Many community models are fine-tuned versions of Qwen.

The Qwen model family

Qwen releases come quickly and the names can be confusing, so here is how the family fits together:

• The Qwen3 series. The current main line, updated through point releases. As of October 2026 the newest is Qwen3.8, which includes a compact model that can run on a single consumer graphics card, a fast and efficient Flash model, and a very large flagship.
• A wide range of sizes. Qwen comes in sizes from small models that fit on a phone or laptop to huge mixture-of-experts models for data centers, so you can pick the trade-off between quality and cost.
• Multimodal models. Recent Qwen models understand images and video natively, not just text.
• Specialist models. Qwen-Image for creating and editing pictures, coding models, embedding models for search, and Qwen Guard models that check content for safety.
• What comes next. Alibaba has said its Flash model previews the design of the coming Qwen4 generation.

Most Qwen models are open-weight, but not all. Some versions are offered mainly through Alibaba Cloud's API, so check the model card for each release.

What Qwen is good at

• Many languages. Qwen is very strong in Chinese and English and supports a long list of other languages, which helps with translation and multilingual products.
• Coding and agent tasks. Recent releases focus on programming, long multi-step tasks and tool use.
• Running locally. The smaller and mid-sized models are a popular choice for running AI on your own laptop or server, especially in compressed versions.
• Fine-tuning and building products. The Apache 2.0 license allows commercial use with few conditions, and common tools support Qwen out of the box.
• Images and documents. The vision models can read screenshots, charts and scanned documents, and Qwen-Image creates and edits pictures.
• Affordable API. Alibaba Cloud offers Qwen models through an API at competitive prices.

Limits to keep in mind

Qwen models can make mistakes and invent details like any AI model, and the smaller sizes are noticeably less reliable than the large ones. Check important facts; our guide on fact-checking AI answers shows a simple routine.

Alibaba's services in China follow Chinese content rules, so the official assistant may avoid some politically sensitive topics. The sheer number of releases and version names also makes it hard to know which model you are using or which one to choose. Running the largest Qwen models yourself needs data-center hardware, and, as with any self-hosted model, adding safety filters is your job.

How to use Qwen officially

You can chat with Qwen for free on the Qwen Chat website (chat.qwen.ai) or in the Qwen app for iPhone and Android. In China, the Qwen app is also connected to Alibaba services, so it can help with tasks such as shopping, food delivery and travel bookings. Developers can call Qwen models through Alibaba Cloud's Model Studio, which uses an API format compatible with OpenAI's. The open weights are published on Hugging Face and on ModelScope, Alibaba's own model hub. The official link in the facts section of this page goes to Qwen's site.

Privacy notes

Privacy depends on how you use Qwen. The international Qwen Chat service is run by an Alibaba company in Singapore. Its privacy policy allows de-identified user content and feedback to be used to improve Qwen's services, including its models, and it describes cross-border data transfers with safeguards. Read the current policy before you share anything sensitive, because it has been updated several times.

If you run the open-weight models on your own hardware, your prompts never reach Alibaba. If you use a cloud host, that host's policy applies. In every case, keep passwords, ID numbers and confidential work data out of public chatbots. Our AI privacy and safety guide explains good habits.

How Qwen compares

Qwen is the most varied open-weight family: if you need a model of a particular size or a permissive license, it is usually on the shortlist. DeepSeek focuses on a few very large, strong models under MIT; Llama has the longest track record but has not had a new generation recently; and Mistral is a European alternative. Closed assistants such as ChatGPT, Claude and Gemini offer more polished apps but no self-hosting. For language practice ideas, see our AI for language learning guide.

FAQ

Is Qwen free?

Qwen Chat is free to use, and most Qwen model weights are free to download. API access through Alibaba Cloud is paid per use.

Is Qwen open source?

Most Qwen models are open-weight, many under Apache 2.0, which allows commercial use. Some versions are only available through Alibaba's API.

Which Qwen model should I run locally?

Start with a small or mid-sized model that fits your graphics card or laptop memory, and move up only if the quality is not good enough. Compressed versions use less memory.

Is Qwen made by Alibaba?

Yes. Qwen is developed by Alibaba's Qwen team and offered through Alibaba Cloud and the Qwen apps.

Is ChatG connected to Qwen or Alibaba?

No. ChatG is an independent AI chat website and is not affiliated with Alibaba. This page is an informational guide. You can try ChatG free in your browser.