Qwen Code is Alibaba's open-source AI coding agent for the terminal and beyond; the CLI is free to install, but model access needs a subscription or API key.
What it is: a multi-surface coding agent from the Qwen team
The repo is Apache-2.0 and written mainly in TypeScript. Docs live at the Qwen Code documentation site. The README's pitch is an agent with Auto-Memory, Auto-Skills, SubAgents, Agent Teams and MCP, open together with the Qwen models, multi-protocol model switching, and clients beyond the terminal.
- Terminal: install,
cd into a project, run qwen, then /auth and a task like explaining the repo. - Desktop: official builds for macOS, Windows and Linux from the
desktop-latest release tag. - Web:
qwen serve --open starts an experimental Web UI. - Editors: VS Code, Zed and JetBrains integrations are documented.
- Chat: Telegram, DingTalk, WeChat and Feishu channels, started with
qwen channel start. - Automation: headless
qwen -p "...", TypeScript / Python / Java SDKs, and an experimental qwen serve daemon over HTTP + SSE.
Install paths: mainline and 1.0 Preview
- Mainline (npm latest 0.27.2 as of November 12): standalone scripts for Linux/macOS and Windows, or
npm install -g @qwen-code/qwen-code@latest (Node.js 22+), or Homebrew brew install qwen-code. Restart the terminal after the standalone install. - 1.0 Preview: released November 11, 2026 as
v1.0.0-preview. It merges agent harness pieces from Qoder with Qwen Code. Install URLs differ from mainline. The preview README says interfaces may still change, community PRs are paused until after December 2026, and issues are still welcome. - npm showed about 178,000 downloads in the week ending November 11. That is the registry's count, not unique users.
How you connect a model
The auth guide's first-run /auth menu has five top-level options. Model access and charges follow the provider you pick.
- Qwen Account with Subscription: Qwen OAuth against a Plus or Pro plan on chat.qwen.ai, for Qwen Code and Agent without pasting an API key.
- QwenCloud: Token Plan or Standard API Key, with an explicit China vs International endpoint pick. Enables the built-in web search tool.
- Alibaba ModelStudio: Coding Plan, Token Plan or a Standard API Key, also with web search on Token Plan and Standard Key.
- Third-party providers: DeepSeek, Grok, MiniMax, Z.AI, Kimi, Idealab, ModelScope, OpenRouter, Requesty and similar key-based presets.
- Custom Provider: OpenAI, Anthropic, Gemini or other compatible endpoints, including local Ollama / vLLM style servers.
Claims, warnings and lineage
- The README compares feature checkmarks with Claude Code and says Qwen Code aims at parity plus extras such as open models and framework, multi-protocol providers, Agent Arena and IM channels. That table is the project's own marketing, not a Seailife review.
- On SWE-bench Verified, the project reports seven Qwen Code versions averaging 76.40%–77.80% with the same Qwen 3.7 Max model (500 tasks, three trials). Configuration details are in their evaluation docs.
- The daemon README warns that by default
qwen serve listens on loopback without bearer auth, so any local process that can reach it can use the full API, including code execution. On shared hosts set QWEN_SERVER_TOKEN and use --require-auth. - The project was originally based on Google Gemini CLI v0.8.2; from Qwen Code v0.1 it stopped syncing and developed independently.
- On November 12 the GitHub repo had about 29,500 stars, about 3,400 forks and about 2,200 open issues and pull requests together. The repository was created on June 26, 2025.
- Seailife did not install the CLI, sign in or run a task. The surfaces, auth options and numbers above come from the README, release notes, auth docs and npm. The logo is the docs favicon; the product image is the official multi-surface banner from the README.