A multimodal large language model (LLM) family developed by Alibaba Cloud. It understands text alongside images, video, audio, and documents, and lets you switch between a “thinking” mode for deep reasoning and a “non-thinking” mode for fast responses. Since its debut in August 2023, the family has evolved through Qwen2 and Qwen3, with many models released as open source under the Apache 2.0 license. You can try it for free through the official chat app (Web/iOS/Android), or download the models from Hugging Face and run them in your own environment.
Key Features
- Multimodal understanding: Handles not only text but also images, video, audio, and documents — from analyzing screenshots to summarizing files, the input range is broad
- Switchable thinking and non-thinking modes: Use the thinking mode for complex reasoning such as math and coding, and the fast non-thinking mode for everyday dialogue, depending on the task
- Open-source releases: Many models, from small ones to large MoE (Mixture-of-Experts) architectures, are published under the Apache 2.0 license, allowing commercial use and fine-tuning
- Strong in agentic and coding use cases: Designed for long-running autonomous tasks and tool calling, it is widely adopted as a base model for coding assistants and agent frameworks
- Multilingual support: The Qwen3 generation is trained on more than 100 languages and dialects, including Japanese
- Official chat app: Qwen Chat (Web/iOS/Android) is free to use after signing up, offering text generation, image understanding, code generation, and more
Pricing
| Usage | Price | Highlights |
|---|---|---|
| Qwen Chat (Web/iOS/Android) | Free | Official chat app with the latest models: conversation, image understanding, code generation |
| API (Alibaba Cloud Model Studio) | Pay-as-you-go | Billed by input/output tokens; rates vary by model |
| Open-source models | Free (Apache 2.0) | Download from Hugging Face and run in your own environment; infrastructure costs are your own |
Pricing is as of August 2026. For the latest API rates, check the official Alibaba Cloud site.
Pros & Cons
✅ Pros
- Many models are open source (Apache 2.0), giving great freedom for commercial use, self-hosting, and fine-tuning
- The official chat app is free, making it easy to try the latest models
- Switchable thinking mode lets you balance reasoning depth against response speed per task
- A wide range of model sizes scales from local execution to full API deployments
- Highly regarded for coding and agentic use cases, with many adoptions as a base model for developer tools
⚠️ Cons
- With so many model variants and generations, beginners may find it hard to choose the right one
- Since the developer is a Chinese company (Alibaba Cloud), some organizations may need to review data handling policies
- Documentation and community resources are mostly in English and Chinese, with little first-party material in Japanese
- Running the open-source models yourself requires substantial GPU resources and operational expertise
Comparison with Similar Services
| Aspect | Qwen | DeepSeek | Llama | ChatGPT |
|---|---|---|---|---|
| Provider | Alibaba Cloud | DeepSeek | Meta | OpenAI |
| Open source | Many models released (Apache 2.0) | Major models released | Released (custom license) | Partial (gpt-oss, etc.) |
| Multimodal | Images, video, audio, documents | Text-centric | Some image-capable models | Images and audio |
| Official chat app | Yes (free) | Yes (free) | Via Meta AI | Yes (free tier) |
| Strengths | Thinking-mode switching, agentic use, multilingual | Reasoning cost efficiency | Ecosystem breadth | Feature range, user base |
Who Is It For
- Developers who want to embed an open-source LLM into their own environment or product
- People looking for a base model for coding assistance or agent building
- Anyone who wants to handle multimodal tasks — images, documents, audio — with a single model family
- Those who want to try the latest models for free through the chat app first
Summary
Qwen is Alibaba Cloud’s LLM family combining multimodal understanding with switchable thinking modes, and its defining trait is that many of its models are open-sourced under Apache 2.0. You can try it casually with the free Qwen Chat, run it in production via the API, or download the models and host them entirely yourself. For developers who value freedom in how they deploy a model, it is one of the strongest options available.