Qwen

Overview

Qwen (Tongyi Qianwen) is Alibaba Cloud's family of large language and multimodal models, first launched in beta in April 2023, together with Qwen Chat, the consumer assistant that runs on top of them. The family spans general LLMs (Qwen3, with hybrid thinking and non-thinking modes), vision-language models (Qwen-VL), coding models (Qwen3-Coder), audio and omnimodal models (Qwen-Audio, Qwen3-Omni), and image and video generation (Qwen-Image, Wan). Many models are released as open weights under Apache 2.0, while flagship variants such as Qwen3-Max are served as a paid API through Alibaba Cloud Model Studio.

What it does

Qwen Chat answers questions and drafts or edits text and code on request. It generates and edits images (Qwen-Image) and generates video (Wan), processes uploaded documents, and grounds answers with web search. Qwen3 can switch between a thinking mode for complex reasoning and a non-thinking mode for faster general chat. DeepResearch plans and runs multi-step web searches and writes a research report, with a dual-mode switch between a faster normal mode and a deeper advanced mode; Alibaba describes it as a multi-agent collaborative mechanism, and it runs under the user's oversight. Qwen Code is a separate open-source coding agent.

Autonomy note

Treat the core Qwen experience as an assistant: it produces output when asked rather than acting on its own. DeepResearch and Qwen Code are the more agentic slices, and they carry out multi-step work under the user's oversight, which makes them supervised rather than fully autonomous.

Integrations & setup

Use Qwen three ways: the free Qwen Chat web app and mobile app at chat.qwen.ai; the pay-per-token API via Alibaba Cloud Model Studio (the platform formerly branded DashScope); or self-hosting the open-weight models, which are distributed via Hugging Face and ModelScope and run with tooling such as Ollama and vLLM. The models support function calling for tool use.

Pricing

Freemium. Qwen Chat offers free access. Open-weight models can be downloaded and self-hosted at no license cost where the Apache 2.0 license applies. Hosted and flagship models are billed per input and output token through Alibaba Cloud Model Studio, with prices varying by model and region (third-party figures are not official); check the official pricing page.

Best for / not for

Best for developers who want strong open-weight models to self-host or call cheaply by API, and for consumers who want a free, broadly multimodal assistant. Less suited to buyers who need a hands-off autonomous worker out of the box; the autonomous slice is limited to DeepResearch and Qwen Code under user oversight, and licensing terms differ across models.

Traction

Alibaba reports having open-sourced over 300 models built on its Qwen and Wan foundation models, with more than 600 million downloads and 170,000+ derivative models created, and the Qwen open models have repeatedly topped Hugging Face open-model leaderboards (vendor-reported and third-party figures).

Alternatives

DeepSeek is the closest competitor (a Chinese open-weight LLM family with a free chat); ChatGPT, Google Gemini, Claude, and Grok are the other leading general assistants. For running Qwen's open weights locally, see Ollama.