# Qwen3-Coder

> Open-source agentic LLM for code reasoning, debugging, and autonomous terminal workflows.

Qwen3-Coder is an open-source Mixture-of-Experts language model built by Alibaba for autonomous coding tasks. It’s trained on 7.5 trillion tokens, mostly code, and activates 35B parameters per task from a 480B total parameter pool. It’s designed for agent-based workflows like pull request bots, debuggers, and file refactorers.

You can use Qwen3-Coder directly in your browser at [chat.qwen.ai](https://chat.qwen.ai) or run it locally via its CLI tool, [Qwen Code](https://github.com/QwenLM/qwen-code). The CLI allows natural language goals to trigger multi-step tasks like reading, categorizing, and sorting files autonomously. It’s available via NPM and requires Node.js.

Qwen3-Coder is compatible with OpenRouter and Alibaba Cloud APIs and is generally 5–7x cheaper than GPT-4 or Claude. The model is Apache 2.0 licensed and optimized for code reasoning, agentic operations, and large codebase understanding.

## Features
- Mixture-of-Experts architecture (480B total, 35B active)
- Trained on 7.5T tokens, optimized for code reasoning and debugging
- Agentic task automation via natural language
- Free web-based chat interface
- CLI for running terminal tasks with goals
- Supports multi-file reasoning and OS-level automation
- API access via Alibaba Cloud and OpenRouter
- Licensed under Apache 2.0

## Integrations
OpenRouter, Alibaba Cloud, Local OS shell (via Qwen Code CLI)

## Platforms
WEB, API

## Pricing
Open Source

## Version
Qwen3-Coder 1.0

## Links
- Website: https://chat.qwen.ai
- Documentation: https://github.com/QwenLM/qwen-code
- Repository: https://github.com/QwenLM/Qwen3
- EveryDev.ai: https://www.everydev.ai/tools/qwen3-coder