Qwen Code changelog: what's new each month
Every stable Qwen Code release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 2 months covered; the current month updates daily.
October 2026 · so far
3 releases: 0.25.0
October brought major collaboration and workspace management improvements to Qwen Code, with new local agent workspace integration, enhanced memory management, and expanded workspace sharing capabilities.
Highlights
- Local workspace-agent collaboration now available with generic Broker provider controls and versioned worker contracts.
- Managed memory now respects configuration settings and uses size-triggered compaction to avoid conversation rewrites.
- Workspace agents gain A2A access and sharing controls for better multi-agent coordination.
- Web Shell gains search within conversations and time-range filtering for trajectory overview.
- Durable remote Shell result delivery and hosted tool approval workflows implemented.
New
- Local workspace-agent collaboration with Broker provider controls for manifest and turn preparation.
- Public Workspace file turns for managed agents (G0).
- A2A access and sharing for workspace agents.
- Hosted tool approval workflows for managed agents.
- Durable remote Shell result delivery.
- Managed runtime attestation client for Java SDK.
- Time-range selection on trajectory overview in Web Shell.
- Search functionality within current conversations in Web Shell.
Improved
- Managed memory availability now respects the memory.enableManagedAutoMemory setting to prevent unwanted memory operations.
- Size-triggered microcompaction clears old tool results and avoids rewriting conversation prefixes on subsequent turns.
- Error messages improved when a skill is invoked as a tool.
- Separate private and group access policies for channels.
- Oversized images returned by MCP tools are now bounded.
- Qwen Live endpoint and model can now be set from settings.
- Qwen Live can run through Web Shell with native macOS Host opt-in.
- Subagent tracking enhanced with parentToolCallId and subagentType for ACP subagents.
Fixed
- Session creation failure diagnostics are now preserved in serve.
- Auth type resolution corrected for ACP agent initialization.
- Subagent text and reasoning chunks now stream correctly in ACP.
- Managed auto memory setting is properly checked before admitting memory requests.
September 2026
31 releases: 0.2.0 → 0.1.17
September brought substantial improvements to the Web Shell interface, expanded workflow and reasoning capabilities, and introduced managed runtime execution foundations. The month focused on enhancing session management, adding Goal automation, and improving integration with external agents and services.
Action needed
- Command hook timeout now read in seconds instead of previous unit.
- Bash will now expand project directory variables in command hooks.
- First-generation Stop-hook Goal implementation removed; use new Goal proposals instead.
- Configurable message-prefix filtering removed from channels; messages follow normal sender and mention policies.
Highlights
- Web Shell now shows live status for running subagents and provides structured shell execution results with timing and token counts.
- Goals can now be proposed by the model and approved in a dialog, then start automatically after their owning turn completes.
- Added trajectory overview showing per-turn durations, time to first token, and tool timing across sessions.
- Introduced managed runtime execution with attestation workers and tool state management for Java environments.
- Web Shell gained search within conversations, time-range selection on trajectory tables, and split-view session navigation.
New
- Standalone chats in Web Shell alongside existing session types.
- QR code button for quick Local Control pairing in Web Shell header.
- Per-run scheduled task sessions that create fresh conversations with unique timestamps.
- Scoped workspace memory tasks with project or user targets and strict filesystem permission boundaries.
- Session turn navigation protocol for bounded turn indexing, snapshots, and transcript paging.
- OpenAI Responses API content generator and support for selecting the OpenAI wire API per model.
- Browser notifications in Web Shell identifying sessions and showing prompt and reply excerpts.
- Playwright-based Browser SDK for expanded browser automation capabilities.
Improved
- Branch picker now displays git state hints like commit count and upstream status.
- Session Overview table optimized with responsive columns, sorting, filtering, and improved actions.
- Managed memory availability now respects the memory.enableManagedAutoMemory setting.
- Size-triggered microcompaction clears old tool results toward a low watermark without rewriting conversation prefix.
- Web Shell composer improved for compactness with clearer context usage across composer and sidebar.
- Live Voice now uses AudioWorklet for microphone capture, reducing main-thread work while preserving metering.
- Built-in websearch enabled by default on ModelStudio Standard and Token Plan.
- Reasoning presets expanded for Kimi, Qwen, and DeepSeek models.
Fixed
- AUTO-mode permission classifier no longer auto-approves out-of-workspace writes; falls back to manual approval.
- Model reasoning preferences now persist in Web Shell.
- Pending permissions and questions now kept across refreshed session loads.
- Document runtime budget headroom restored during export.
- Responses stream error details now preserved in core.
- Node-pty's conout worker released after every PTY on Windows.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Qwen Code: pricing, features and alternatives · All changelogs