AI Briefingqwen-code: Open-source terminal coding agent that achieves 77.8% on SWE-bench
QwenLM/qwen-code
About the project
An open-source AI coding agent that runs in the terminal. Both the Qwen model and the framework are open-sourced, allowing usage without vendor lock-in. It recorded a score of 77.80%, up from an average of 76.40%, on the SWE-bench Verified benchmark when combined with the Qwen 3.7 Max model.
Developed with the goal of functional equivalence to Claude Code, it natively supports sub-agents, automatic memory, MCP, and LSP integration. It features a multi-protocol structure that allows free switching at runtime between OpenAI, Anthropic, Gemini, and Qwen APIs, as well as local models like Ollama and vLLM.
Beyond a simple terminal tool, it provides IDE plugins for VS Code, Zed, and JetBrains, along with a desktop app and web UI. It also supports integration with messenger channels such as Telegram, DingTalk, WeChat, and Feishu, enabling agent sessions to be shared across various environments.
It can be integrated into CI/CD pipelines or scripts for automation tasks via TypeScript, Python, and Java SDKs and headless mode. When run in daemon mode, browsers, editors, and automation clients share and work on the same agent session via HTTP and SSE.
QwenLM/qwen-code
An open-source AI coding agent that lives in your terminal.
TypeScript
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.