Qwen Code v0.21.10 Changelog

This release is all about making the software more stable and pleasant to use, especially if you use the Web Shell (the browser-based interface). It fixes a lot of visual glitches and bugs that could freeze the app, while adding a few handy features like better image previews and smarter AI reasoning controls. This matters to anyone who uses this tool daily for coding, as it saves time and prevents frustration.

What's new

• Adjustable AI thinking effort: If you connect to this tool using a client program (via something called ACP), you can now tell the AI exactly how hard to think about a problem—ranging from "Default" up to "Max."
• Image previews: When you paste or upload a picture into the Web Shell, you can now click on it to see a full-size preview in a side panel and download it if you need to.
• Automatic log cleanup: The app now automatically deletes old log files when it's running in the background (like "headless" or "daemon" modes, which just means running without a user interface). This prevents your hard drive from filling up with junk data.
• Faster session loading: The Web Shell now shares saved (cached) data between different parts of the app, so your list of previous chat sessions loads much faster and wastes less internet data.

Improvements

• Smoother Web Shell: Scrolling through really long chat histories is now stable and won't jump around or lose your place. Animations are less stuttery, and the timer that shows how long a task has been running no longer randomly resets.
• Better stats for big files: When the AI edits huge files, the Web Shell used to struggle to count how many lines changed. It now uses a smarter, faster math method (a "linear-time algorithm") to show you accurate line-change numbers without freezing.
• Faster command-line performance: The main text-based app (the CLI) now saves a temporary list of your sessions. This makes it much faster to organize and filter your background tasks, and it automatically updates the list whenever something changes.

Fixes

• Fixed AutoFix freezing: "AutoFix" is a feature that automatically fixes code errors. It used to freeze and time out during really long background tasks because the system thought it had stopped working. It now streams live progress updates so the system knows it's still busy.
• Fixed Windows startup crash: The desktop app was completely failing to open on some Windows computers because it got confused by how Windows formats file locations (specifically, "verbatim path prefixes"). This is now fixed.
• Fixed enterprise network blocking: The desktop app's local control feature used to block connections from company networks (enterprise LANs) because it was too strict about what internet addresses were safe. It now safely allows these company network addresses.
• Stopped duplicate submissions: If you refreshed your browser while the AI was in the middle of typing out an answer, it used to accidentally submit your prompt twice. It now syncs properly with the background system to prevent this.
• Fixed screen flickering: Resizing your terminal window or waking your computer up used to cause the app's banner text to duplicate and the screen to flicker. That visual glitch is gone.
• Fixed broken Chinese web links: If you were chatting in Chinese, web links (URLs) in the AI's answers would sometimes break because the system didn't know how to handle Chinese punctuation marks next to the link. It now correctly stops the link before the punctuation.
• Simpler menu navigation: When using the @ menu to mention things, you no longer have to press awkward keyboard combos like Ctrl+Arrow to switch categories. You can just use the regular Left and Right arrow keys.
• Fixed missing tools after resuming: If you paused a session and came back to it later, some special background tools (called MCP tools) wouldn't show up. They now reappear correctly when you resume your work.
• Fixed session retry lock-out: If a session took too long to load and timed out, the app used to get confused and refuse to let you try loading it again. It now clears out the old, stuck ID so you can easily retry.
• Removed duplicate UI clutter: The "context usage" bar (which shows how much of the AI's memory you are using) used to show up twice on your screen if it was already in the status line. Now it only appears once.
• Fixed data loss when switching sessions: Switching between different chat sessions in the web interface is now much safer. It uses a "transactional" system—meaning if a session fails to load, it won't corrupt your data or erase your progress.
• Fixed cut-off tables: If you pressed Ctrl+S in the command-line app to save your history, long tables and text were getting chopped off. They now fully expand so you can see everything.
• Smarter automatic titles: When the app automatically names your chat sessions, it used to get confused and name them after boring background code ("hook context") instead of what you actually asked. It now correctly names the session based on your real prompt.
• Fixed confusing stop markers: The system uses visual markers to show when an AI task stops. These markers were acting glitchy and giving contradictory information about whether a task was actually done. They now clear out properly.
• Fixed model version mix-ups: The system was confusing the version of an AI model you have installed with the template used to talk to it. It now correctly tells them apart.
• Better error codes for closing: When a session closes, it now uses a structured error code. This sounds technical, but it just means the Web UI and Java clients can reliably understand what happened and automatically try again if needed.
• Restored background downloads: The system that sends out new live updates to a cloud storage service (Aliyun OSS) was broken because the permissions were wrong. This is fixed so updates can mirror properly again.

Breaking changes

No breaking changes in this release.