Qwen Code v0.21.0 Changelog
What changed in Qwen Code CLI v0.21.0, released 2026-07-24 โ explained in plain English.
- The web interface got a major upgrade, making it much easier to manage multiple projects and preview files.
- The automated Pull Request (PR) system is now smarter about fixing itself when it gets stuck or crashes.
- You can now teach the AI using videos, and it will remember past chat sessions if you tag them.
- Several annoying visual glitches and security quirks were fixed to make the tool safer and smoother.
Want a heads-up when Qwen Code updates?
We email you the moment a new Qwen Code version drops โ plus plain-English release notes like these. No spam.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
This release is all about making the AI coding assistant more independent and easier to manage. It matters most to developers who use this tool to automatically review and fix their code, as well as teams who use the web interface to keep track of multiple AI agents working at the same time.
What's new
- Better Web Workspace Management: You can now easily switch between different projects (workspaces) and manage background AI agents directly from the web interface.
- File Previews: The web interface can now show you a visual preview of files instead of just text.
- Smarter Auto-Fixes: If the AI tries to submit a code fix (a Pull Request) and it fails to upload multiple times, it will now stop trying so it doesn't get stuck in an endless loop. It will also automatically try again if a system crash interrupted it.
- Video Learning: You can now feed native video files into the
/learncommand to teach the AI new things. - Tagging Past Chats: You can now use the
@symbol to reference previous conversations, making it easier to bring up old context. - Custom AI Identity: Developers can override the AI's core personality or instructions using a specific file setting.
- Bilingual Code Reviews: If you write your Pull Request description in Chinese, the AI will automatically post its code review in both Chinese and English.
- Configurable Image Models: You can now choose which specific AI model to use for generating images.
Improvements
- Background Agents: Completed background agents are now kept around so you can review what they did, and their task revival is now clearly advertised.
- Cleaner Web Styling: The web interface now uses "Shadow DOM" (a web standard that keeps code isolated). This means the tool's visual style won't accidentally clash or mess with the website it's embedded on.
- Safer Background Tasks: The system now strips away secret internal passwords from background tasks, making it more secure.
- Smoother Terminal Links: Login links in the terminal now use a standard format (OSC 8) that lets you click them directly, instead of copying and pasting a long URL.
- Smarter Scheduling: The system's internal timer (which works like a cron job, a scheduled task) was updated to follow standard scheduling rules more accurately.
Fixes
- Flickering Text: Fixed an annoying flickering issue with the AI's "thinking" text while it was typing, and made Shift+Enter correctly create a new line.
- Redundant Checks: The web interface stopped making unnecessary, repeated requests to check your Git status (saving computer power).
- Empty Folder Crashes: Fixed a bug where the AI would crash if you gave it an empty folder path when it was expecting a specific working directory.
- Port Detection: Fixed an issue where the tool wouldn't open the correct backup port if the main one was already taken.
- Image Support: Fixed a bug so that specific newer AI models (like qwen3.8-max and kimi-k3) can properly accept and understand images.
Breaking changes
No breaking changes in this release.
How to upgrade
Update to Qwen Code v0.21.0 by running:
npm install -g @qwen-code/qwen-code
This is our plain-language summary. Read the complete, official notes on the Qwen Code official changelog โ.
