Bug Fixes Fixed an issue where sub-agent model groups displayed inconsistently and did not refresh upon re-entry.
New Features Added free early claiming of Weekend Plan — invite friends to claim it together Plugins referenced in prompt templates now support one-click installation, with clearer progress indicators Added a skills shortcut to the input box menu Thinking traces now support content search, with more complete records and expanded by default MCP servers now support configuring the protocol version, with guidance prompts when negotiation fails Edited messages can now be sent immediately without waiting in the queue View an execution summary and duration when a conversation turn ends The "Browser" menu in Settings has been renamed to "Browser Control" Bug Fixes Improved the login authorization flow and fixed occasional callback failures Fixed an intermittent mcp oauth failure Model responses that occasionally cut off or come back blank now retry automatically Fixed an issue where queued messages were lost after an error and interrupted task statuses were displayed incorrectly Unified tool execution status wording; thinking process labels are now more concise and clearer Fixed issues where task action hints were clipped and task information was jumbled on touchscreens Unread markers are now cleared immediately after opening the task timeline Raised the threshold for converting pasted long text into attachments, reducing unintended conversions Fixed an issue where a single tilde was incorrectly displayed as strikethrough The input box no longer loses focus when switching sidebars Input box references and command menus are displayed more clearly, and search no longer retries repeatedly Fixed an issue where multiple skill directories in a workspace could not be listed at the same time Fixed issues where some plugin skills were not recognized and their names displayed inconsistently Usage statistics now show the last seven days by default, with clearer charts and value labels Fixed an issue where background processes remained after exiting on Windows Fixed several issues with Computer Use on macOS and in remote workspaces Improved browser control session stability; page script execution now behaves as expected Fixed an issue where the remote interface menu overflowed the screen on mobile Fixed an issue where model status displayed incorrectly after a plan expired Fixed an error when modifying model configuration while a task is running
New Features Added the GLM 5.3 Flash multimodal model, available out of the box for subscription users. Improved the Computer Use experience, with more accurate recognition and more stable operations when paired with GLM 5.3 Flash. Required permissions are now clearly prompted before using the computer control feature. The computer control feature is now supported on macOS models with Intel chips. Browser operations now preferentially reuse existing tabs. Bug Fixes Fixed an issue where tabs would occasionally freeze or fail to restore during browser operations. Fixed incorrect plugin name display; names are now consistent with the plugin marketplace. Fixed an issue where computer control operations would occasionally stop working on macOS.
New Features Optimize the model configuration options page By default, the thinking process and real-time time consumption are displayed Issue Fix Fixed the issue where screenshots and web tool images could not be displayed in some gateways Fixed the issue where single-colon file references could not be recognized
New Features Support global and workspace-level management of agent capabilities, with visual management in Settings Add recommended prompts on the home page, with support for referencing enabled plugins Add a 5-hour quota reset benefit during off-peak compute hours for GLM Coding Plan subscribers Upgrade usage statistics to support Personal Plan V3 and Team Plan usage tracking Add retry failure pages and custom timeout support for the project knowledge base Improve file references, code comments, and tool call grouping in conversations Add video recording support to the built-in browser Automatically preview generated presentations and refresh previews after edits Support secure workspace-level Hooks configuration and execution Upgrade the built-in document plugin Bug Fixes Improve stability for project knowledge base generation, retry, and failure recovery Fix built-in browser startup failures, crashes, and leftover tabs Fix side terminal interruptions or input failures after switching workspaces Fix Windows IME input issues in the side terminal Fix white flashes when refreshing the task list and abnormal file tree scrolling Fix references disappearing after typing spaces following plugin trial usage Fix service status not updating correctly after refreshing the Settings page Fix occasional failure to save window size when exiting the app Fix duplicate usage counting, chart clipping, and display issues in usage statistics Fix external link right-click open behavior and multiple Chinese/English display issues Fix code review comment positioning and preview status issues Optimize startup loading speed for very large conversation histories Fix delayed remaining usage updates
New Feature The brand-new flagship model GLM-5.3 is now available! Bug Fixes Fixed an issue where reasoning levels did not take effect in some automated tasks. Fixed an issue where repository knowledge base generation could not use reasoning levels. Fixed an issue where automatic git commit message generation failed due to missing reasoning levels.
New Features You can now comment on individual selected PowerPoint elements with clearer guidance. Headless sessions now skip memory extraction for leaner unattended runs. Bug Fixes Conversations now restore compacted usage details correctly after restarting the app. The app can now recover automatically from a corrupted task index during startup. Remote sessions now resend missed updates after reconnecting. Automation cancellation, limits, and partial creation states now behave more reliably. Provider quota errors now stop retrying instead of repeating failed attempts. AI responses and usage tracking are now more consistent across retries and model changes. Public GitHub plugin sources can now install correctly even without Git.
New Features Memory now supports browsing by project and viewing project memories in Settings. The task workspace now supports grouped task sections. @ mentions now support plugins, files, and conversations. Idle tasks now support subagents configured with custom models. Automation tasks now support custom minute-based intervals. Subagents now support a reasoning effort setting. Bug Fixes Improved memory navigation, icons, actions, and preview stability. Fixed conversation jump positioning after history changes. Fixed preview and positioning issues when dragging grouped project tasks. Improved layout stability for conversations, status panels, long task titles, and side tabs across window sizes. Improved readability for search history, timestamps, working duration, and helper text. Improved document, presentation, and file preview rendering and refresh behavior. Fixed preview issues caused by cached image attachments. Fixed incorrect automation task model and target display. Fixed queued messages that could be reordered or sent too early. Fixed model selection loss after switching runtime environments. Fixed blank or missing conversations after reconnecting to remote workspaces.