Qwen

Qwen is the AI made by Alibaba's Qwen team: a free chat assistant at chat.qwen.ai and in apps, and one of the largest families of open-weight models. Its newest generation is Qwen3.8, released in Aug 2026, led by Qwen3.8-Max, a 2.4-trillion-parameter model that Alibaba calls the first Max-class Qwen it has released openly, and Qwen3.8-27B, a compact model under the Apache 2.0 license (as of Nov 6, 2026).

Qwen is the pick for developers who want open models in every size, from phone-sized to frontier-sized, with permissive licenses, and for a capable free chatbot with image and video generation. For a polished assistant with the most tools and integrations, ChatGPT, Claude or Gemini are more complete. Alibaba is a Chinese company; see Privacy below. Compare chatbots in Best AI Chatbot.

Quick facts

Features

Qwen Chat. The free chat app covers chatting, image and video understanding, image generation, document processing, web search, tool use and Artifacts, as its own description lists them. Models can be picked in the app, including Qwen3.8-Max.

Thinking control. Qwen3.8 models can think before answering, and developers set how deeply with a reasoning effort setting, per the model cards.

Vision and video. Qwen3.8-27B is natively multimodal, reading images, documents and videos up to hours long, according to Alibaba's model card.

Image models. The Qwen team also publishes image generators, most recently Qwen-Image 2.1 (Sep 2026).

Models and licenses

Sources: model cards for Qwen3.8-Max, Qwen3.8-Flash-Next and Qwen3.8-27B, read Nov 6, 2026.

The Qwen3.8-Max License allows commercial use, with two conditions: a product with more than 100 million monthly users or $20 million in monthly revenue must show the model's name, and a company selling model-as-a-service or an AI coding or office assistant with more than $50 million in yearly revenue needs a separate license. Qwen3.8-27B under Apache 2.0 has no such conditions, and is the size most people can Run AI Locally.

1:08 Smaller Qwen models download and run on your own computer through Ollama or LM Studio. In a Mar 18, 2026 video, Nick of Zero to MVP ran Qwen 3.5's 9-billion-parameter model that way on his laptop, because the 27-billion-parameter version needed more video memory than it had.

3:25 In his tests, the local 9-billion-parameter model failed to build a web page of sorting algorithms after more than half an hour of rewriting its own code, but correctly added a working copy button to a small app.

7:20 Given the same sorting task, Qwen 3.5's largest model, with 397 billion parameters and run in Ollama's cloud, built a working page on the first try, which he judged able to compete with paid models from major companies.

History

• Apr 29, 2025: Qwen3, eight open models from 0.6 billion to 235 billion parameters, all under Apache 2.0.
• 2026: the Qwen3.5 and Qwen3.6 series, which Alibaba says saw wide community adoption.
• Aug 2026: Qwen3.8, including the first openly released Max model.
• Sep and Oct 2026: Qwen-Image 2.1 and its Turbo version.

Alibaba has not published a current user count for the Qwen app on the pages read for this article.

Privacy and data

Qwen Chat is run by Alibaba, a Chinese company; read its privacy policy before sharing sensitive material. The open weights avoid the question entirely: downloaded and run on your own machine or a cloud you choose, nothing goes to Alibaba. See AI Privacy Compared.

Qwen vs the alternatives

DeepSeek uses the MIT license on all its weights and has very low API prices. Kimi and GLM are strong Chinese open models for coding agents; Zhipu's GLM-5.3 card compares itself with Qwen3.8-Max directly. Llama and Gemma are the US open-model families, and Mistral the European one. Choose Qwen when you need many sizes from one family, especially the Apache 2.0 27B model.

Frequently asked questions

Is Qwen free? Yes. Qwen Chat at chat.qwen.ai is free, and Qwen's open models can be downloaded free from Hugging Face; Alibaba charges only for API use through Qwen Cloud.

What is the latest Qwen model? Qwen3.8, released in Aug 2026; its largest model is Qwen3.8-Max, with 2.4 trillion parameters and open weights.

Is Qwen open source? Most Qwen models are open weight. Qwen3.8-27B uses Apache 2.0; Qwen3.8-Max uses its own license, which adds conditions only for very large products and model-as-a-service companies.

Who makes Qwen? Qwen is made by the Qwen team at Alibaba, the Chinese e-commerce and cloud company.

Can I run Qwen on my own computer? Yes. Smaller Qwen models, such as Qwen3.8-27B, run with tools like Ollama or LM Studio on a computer with a strong GPU; the 2.4-trillion-parameter Max needs a server cluster.

Sources

• Qwen3.8-2.4T-A95B and its license, Qwen on Hugging Face, read Nov 6, 2026
• Qwen3.8-27B, Qwen on Hugging Face, read Nov 6, 2026
• Qwen3.8-Flash-Next, Qwen on Hugging Face, read Nov 6, 2026
• Qwen models, Hugging Face, read Nov 6, 2026
• Qwen3.8-Max, Qwen Cloud, read Nov 6, 2026
• Qwen Chat, Alibaba, read Nov 6, 2026
• Qwen 3.5 Locally: Can It Replace Paid AI Models?, Zero to MVP, Mar 18, 2026 (video)
• Qwen3: Think Deeper, Act Faster, Qwen, Apr 29, 2025