Qwen Code: what it is and which models you can run in it

The query "what is Qwen Code" appears because the name sounds like several different things: the Qwen Coder model, the Qwen chat and a terminal program. Here it is briefly and to the point: what Qwen Code is, how it differs from the model, how to start it and which models you can plug in. The description relies on the project's README and documentation, checked as of October 18, 2026.

Qwen Code vs Qwen Coder

These are two different things.

• Qwen Coder (the Qwen3-Coder family) is a set of language models from the Qwen team tuned for code. A model on its own just receives text and returns text.
• Qwen Code is an agent program. It runs in the terminal, reads your project, proposes edits and executes commands, while the connected model does the thinking. The model in it can be swapped.

The easiest comparison is Claude Code: it is also a program, and Claude inside it is the model. According to the README, Qwen Code was originally built on Google Gemini CLI (version 0.8.2) and has been developed independently as a multi-protocol agent since version 0.1.

How the agent works

Every agent runs the same loop: you write a task, the program collects context (project files, instructions, conversation history), sends it to the model, receives the reply and, if the model asked to call a tool (read a file, run a command, edit), runs it and sends the result back. This repeats until the task is done. More on the general principle: what an agent harness is↗.

According to the project README, Qwen Code has:

• subagents, skills, hooks, MCP, a planning mode and a sandbox;
• memory: a QWEN.md file with your instructions and automatic agent notes;
• a headless mode for scripts and CI: qwen -p "...";
• SDKs for TypeScript, Python and Java;
• plugins for VS Code, Zed and JetBrains, plus a desktop app and an experimental web UI (qwen serve --open).

Some features are marked as beta or experimental in the docs, so check their current status there.

How to start

You need Node.js 22 or newer if you install through npm:

npm install -g @qwen-code/qwen-code@latest
cd /path/to/project
qwen

The README also offers an install script for Linux and macOS, a Windows version and Homebrew (brew install qwen-code). After the first launch run /auth inside the session and choose how to connect a model. The free Qwen OAuth login, according to the docs, was discontinued on April 15, 2026, so you need your own key: details are in the section on authentication in the Qwen Code docs.

Which models you can use

The main feature of Qwen Code is that it is not tied to one model. The documentation lists several protocol families:

In /auth there are built-in flows for the Alibaba cloud, for several third-party providers and a "Custom Provider" entry for any address of your own. Models are switched with /model right in the session.

Models available through Zukko

Through the OpenAI-compatible gateway Qwen Code can run any model from the catalog. Prices per 1M tokens as of October 18, 2026:

The strength index is the Intelligence Index from Artificial Analysis; we give no other quality figures. Which model handles your code best only a test on your own tasks will show. The current list: GET https://api.zukko.pro/v1/models or the "Models" page in the dashboard.

Minimal setup for Zukko

export OPENAI_API_KEY="fetch_your_key"
export OPENAI_BASE_URL="https://api.zukko.pro/v1"
export OPENAI_MODEL="qwen3.8-max"
qwen

Create the key on the API keys page↗. A persistent setup through ~/.qwen/settings.json and a model list is described in Qwen Code CLI: connecting↗.

An example cost (hypothetical volumes): a session with 500K input tokens and 30K output. On qwen3.8-flash: 0.5 × 0.038 + 0.03 × 0.075 = 0.019 + 0.00225 = $0.02125; on qwen3.8-max: 0.5 × 0.088 + 0.03 × 0.238 = 0.044 + 0.00714 = $0.05114. Input in an agent grows with every step because the context is resent, and for Qwen models in Zukko caching does not lower the total: a cache read costs the same as input.

Who it is for

Qwen Code fits if you:

• work in the terminal and want an open agent;
• want to change models without changing the tool;
• want to use the agent the same way locally and on a server over SSH.

If you need an agent right in the editor, see the overview of VS Code extensions for AI↗ and the article Qwen for development↗.

FAQ

Is Qwen Code the same as Qwen Coder?

No. Qwen Coder is a model, Qwen Code is an agent program that can use that model or any other.

Does Qwen Code only work with Qwen models?

No. According to the docs it supports OpenAI-compatible, Anthropic and Gemini protocols, as well as local servers such as Ollama and vLLM.

Do I need an Alibaba account for Qwen Code?

Not necessarily. You can supply a key from any compatible service, including a gateway key. You can top up the balance from Russia by SBP, MasterCard or crypto.

Does Qwen Code have a graphical interface?

Yes: a VS Code plugin, Zed and JetBrains support, a desktop app and an experimental web UI. See the project README for the list and status.