QwenQwen Code is an open-source coding agent that runs in the terminal. It ships with automatic memory and skills, subagents, agent teams, and MCP support, and it speaks several provider protocols, so the same session can run against OpenAI, Anthropic, Gemini, or Qwen APIs, a third-party provider, or a local model, switched at runtime.
Beyond the interactive terminal UI, it runs headless for scripts and CI with qwen -p, as IDE plugins for VS Code, Zed, and JetBrains, as a desktop application for macOS, Windows, and Linux, as a daemon (qwen serve) that shares one agent session with multiple clients over HTTP and SSE, through TypeScript, Python, and Java SDKs, and as a chat bot on Telegram, DingTalk, WeChat, and Feishu.
The project began from Google's Gemini CLI v0.8.2 and stopped syncing with upstream at v0.1, developing independently as a multi-protocol, multi-platform agent framework. Published evaluation runs use SWE-bench Verified over 500 cases, three trials per version across seven versions, with Qwen 3.7 Max at a one-million-token context and a 500-iteration harness limit.
Features
- Agentic defaults: automatic memory and skills, subagents, agent teams, dynamic workflows, and MCP with no setup step
- Multi-protocol models: OpenAI, Anthropic, Gemini, and Qwen APIs, third-party providers, and local models, switchable at runtime
- Run modes: interactive terminal UI, headless qwen -p, IDE plugins, desktop app, daemon mode, SDKs, and instant-messaging bots
- Built-in skills: slash commands including /review, /batch, /loop, and /bugfix, plus plan mode, hooks, and LSP integration
- Sandboxing and worktrees: auto mode, a sandbox, and git worktrees for isolated runs
- Agent Arena: several models run head-to-head on the same task
- Computer use: desktop automation alongside code editing
- Localised docs: documentation in Chinese, German, French, Japanese, Russian, Brazilian Portuguese, and Korean
| Version | Avg score | Pass rate | pass@3 | pass^3 |
|---|---|---|---|---|
| 0.0.14 | 77.80% | 84.6% | 84.6% | 69.2% |
| 0.1.0 | 76.73% | 84.6% | 84.6% | 68.0% |
| 0.5.0 | 76.60% | 83.0% | 83.0% | 68.4% |
| 0.10.0 | 76.47% | 84.4% | 84.4% | 67.4% |
| 0.15.0 | 77.67% | 84.6% | 84.6% | 68.6% |
| 0.19.0 | 76.40% | 85.0% | 85.0% | 65.8% |
| 0.22.0 | 77.33% | 84.4% | 84.4% | 68.0% |
Integrates with
AionUi
Free, local, open-source 24/7 Cowork app for OpenClaw, Hermes Agent, Claude Code, Codex, OpenCode, Gemini CLI and 20+ more CLI.
-
Ollama
Runs open large language models locally without an API key
vLLM
LLM inference and serving library using PagedAttention and continuous batching, with an OpenAI-compatible API server
Integrated by
Agent Shell
A native Emacs buffer to interact with LLM agents powered by ACP
Context Mode
Your coding agent loses track of what it was doing after 30 minutes because every tool call dumps raw output into the co
jCodeMunch MCP
Cut AI token costs 95%+ on code exploration. The leading MCP server for precise, symbol-level GitHub code retrieval via tree-sitter AST. Wor
TokenTelemetry
Token telemetry dashboard for AI autonomous and coding agents — tracks tokens, sessions, tool calls & reasoning across Hermes agent, Claude
