Alibaba's terminal AI agent plays nice with Claude and GPT too

What it does Qwen Code is a terminal-first AI agent that reads, refactors, and generates tests for large codebases. While optimized for the Qwen3-Coder model family, it speaks OpenAI, Anthropic, and Gemini protocols, so you can point it at Alibaba Cloud, OpenRouter, or your own API endpoint. Optional IDE extensions for VS Code, Zed, and JetBrains exist for when you leave the command line.

The interesting bit The project open-sources both the agent framework and the Qwen3-Coder model, shipping updates to both in tandem so the tooling and the brain do not drift apart. It also exposes a Skills and SubAgents system that breaks complex tasks into smaller, delegated workflows rather than monolithic prompts.

Key highlights

• Multi-protocol provider support: OpenAI-compatible, Anthropic, and Google GenAI endpoints all configurable in ~/.qwen/settings.json
• Agentic workflow with built-in Skills and SubAgents for decomposing tasks
• Terminal-native with optional IDE integrations for VS Code, Zed, and JetBrains
• Both the agent framework and the underlying Qwen3-Coder models are open-source
• Supports thinking mode for models that offer it, such as qwen3.5-plus

Caveats

• The free Qwen OAuth tier was discontinued in April 2026; you now need an API key or an Alibaba Cloud Coding Plan subscription
• Requires Node.js 22 or later
• Documentation and configuration examples heavily favor Alibaba Cloud endpoints, so using other providers means studying the JSON schema

Verdict Developers who want a Claude Code-style terminal agent but need to switch between LLM providers—or prefer open-source tooling that ships alongside its own model—should look here. If you need a fully offline, local-only coding agent, keep looking; this tool is API-dependent and currently nudges you toward Alibaba Cloud billing.