Qwen Code

Qwen Code is an open-source AI coding agent that runs in the terminal, in editors, as a desktop app, in a browser, and in chat apps. It began as a fork of Google's Gemini CLI and has developed independently since version 0.1 as a multi-protocol agent framework with deep integration for Qwen models.

It speaks the OpenAI, Anthropic, Gemini, and Qwen APIs, works with third-party providers and local models served by Ollama or vLLM, and switches model at runtime. Agent features include auto-memory, auto-skills, subagents, agent teams, dynamic workflows, hooks, MCP, plan mode, LSP integration, a sandbox, Git worktrees, and computer use, with a stated goal of feature parity with Claude Code.

Beyond the interactive terminal, Qwen Code runs headless in scripts and CI/CD, through TypeScript, Python, and Java SDKs, and as an experimental daemon that serves several clients over HTTP and server-sent events using ACP. Other agents such as Claude and Codex can delegate coding tasks to it through Acpx.

On SWE-bench Verified (500 cases, three trials per version, with Qwen 3.7 Max), seven Qwen Code versions from 0.0.14 to 0.22.0 score average results between 76.4% and 77.8%, with pass rates between 83.0% and 85.0%.