Qwen-Code v0.12.5 Preview Released

🔑 Enhanced Key Takeaways

• •Qwen-Code v0.12.5-preview introduces Plan Mode, featuring a pre-execution planning pane that requires explicit user approvals for file changes, operations, and rationales to enhance safety[1][2].
• •Automatic Vision Intelligence integration switches to Qwen3-VL-Plus model upon detecting images, supporting 256K input and 32K output tokens for extended context handling[1].
• •Recent updates include Zed editor OAuth support for both OpenAI and Qwen providers, resolving authentication hangs and improving multi-provider compatibility[1].

🛠️ Technical Deep Dive

• •Built as a fork of Gemini CLI, incorporating smarter defaults and compatibility for downstream tools without manual model switching[1].
• •Includes subagent performance enhancements, high-resolution image support for Qwen3-VL-Plus, loop detection toggle, and safeguards against overwriting existing QWEN.md files[1].
• •Windows-specific fixes cover multi-line paste handling, Markdown list rendering, malformed tool calls, and output token limits[1].

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline