Qwen Code v0.21.5 Changelog

This release is all about making the app smarter, safer, and less frustrating to use. It matters most to developers and power users who rely on Qwen Code to write, review, and manage their software projects. Along with a helpful upgrade path for Mac users, the update focuses heavily on fixing bugs that caused the app to waste time or get stuck, while adding clearer ways to see how the AI is thinking.

What's new

• Mac Upgrade Bridge: Mac users now have an optional, one-time helper to move from the older "Electron" app (a way to build desktop apps using web tech) to the newer "Tauri" app (a faster, lighter way to build desktop apps).
• Better Tool Tracking: When the app uses a tool to do something, it now keeps a detailed record of the exact outcome—whether the tool worked, failed, or was canceled.
• Clearer Code Reviews: The Web Shell (the browser version of the app) now shows code review results in a neat, organized way. It highlights how bad a problem is (severity), how sure the app is about it (confidence), and suggested fixes.
• Adjustable AI Thinking Effort: The app can now support different "reasoning effort levels" for the new Qwen 3.8 model, meaning you can tell the AI how hard to think about a problem based on your settings.

Improvements

• Faster internal testing: The developers sped up their own background tests by canceling old, outdated test runs and sending certain jobs to faster computers. This helps them get new features to you quicker.

Fixes

• Endless Goal Loop: Previously, if the app ran out of "evidence" (information it uses to prove a task is done) in Goal mode, it would get stuck trying the same thing over and over. It now stops and asks you for help instead of looping forever.
• Wasted Review Runs: The review system used to re-run its checks even on tasks it had already successfully finished. This is fixed, which saves time and computing power (tokens).
• Unsafe Tool Replays: If your connection dropped while the app was using an MCP tool (a plugin that lets the AI talk to outside software), it used to blindly try the action again. This could be dangerous! Now, it will only automatically retry an action if the tool is specifically marked as safe to repeat ("idempotent") and you trust the workspace.
• Read-Before-Write Rule: The app's instructions were updated so that it must always read a file before writing to it if it doesn't know what's already in there. This prevents accidentally overwriting important data.
• Table Pop-up Glitch: Closing a pop-up window while editing a Markdown table used to make the page jump around. This is fixed, and your keyboard focus stays exactly where it should be.
• Test and Workflow Fixes: Several behind-the-scenes tests were fixed, including reading logs from the correct folder on Windows and fixing a broken system that sorted incoming bug reports.

Breaking changes

No breaking changes in this release.