Follow us on LinkedIn, X, Bluesky, Instagram | Follow Insiders Changelog on X or Bluesky
Last updated: September 21, 2026
Welcome to the 1.139 Insiders release of Visual Studio Code.
These release notes cover the Insiders build of VS Code and continue to evolve as new features are added.
You can still track our progress in the Commit log and our list of Closed issues.
These release notes were generated using GitHub Copilot and might contain inaccuracies.
Happy Coding!
September 20, 2026
- Open Keyboard Shortcuts and Running Extensions as modal editors in the Agents window, so these utility editors no longer replace the active session. #336936: Open utility editors modally in Agents Window
September 18, 2026
-
Rename Preview settings for completed agent sessions from Archive to Mark as Done, and migrate existing automatic-cleanup values to the renamed settings. #335801: Rename Archive settings to Mark as Done
-
Add
**/.jj
to the default files.exclude patterns so Jujutsu repository metadata stays hidden in the Explorer. #308943: Exclude .jj (Jujutsu VCS) directory by default -
Add model picker controls for choosing a model's Thinking Effort and, when offered by the model, a longer Context Size. #280436: Feature request: Configurable reasoning level and "max context" mode in VS Code Copilot Chat
-
Preserve an explicit Mark as Unread state for the focused agent session until you leave and return to the session or mark it as read. #336721: Mark as Unread does not mark the currently focused session as unread
-
Remove stale persisted references and artifacts from agent sessions without deleting the underlying file, issue, or pull request. #336500: Remove stale Agent Host session references
September 17, 2026
- Add an Optimize for control to the Auto model, with Efficiency, Balance, and Intelligence options for balancing model capability, speed, and cost in Copilot Chat and Copilot SDK agent sessions. #333266: Support for Auto Tiers
September 16, 2026
-
Update the desktop runtime to Electron 43.6.0, improving VoiceOver Braille line tracking on macOS and native window controls on Linux. #331466: Electron 43 update
-
Add keyboard shortcut to mute or unmute the microphone while experimental Voice Mode is connected and the chat input is focused. #336280: Easier mute/unmute for GitHub Copilot Voice Mode
September 15, 2026
-
Add Cargo completions to terminal suggestions, including subcommands, options, packages, targets, features, crates, toolchains, and plugins. #336216: Terminal suggest: Add Cargo completions
-
With experimental Agent Merge enabled, agents can turn it on for the current session, so you can ask an agent to create a pull request and take it through review and CI. #334033: Agent should be able to enable agent merge on a PR
September 14, 2026
-
Add inline color previews for
lab()
andlch()
color values in CSS. #165207: [css] Add new CSS color functions -
Extension authors can use
display:inline-block
in sanitized Markdown span styles to size background highlighting without relying on Codicon classes. #180496: Allow display prop on markdown span style
September 11, 2026
-
Make long Remote workspace menus in the Agents window easier to navigate with the mouse by making them searchable and scrollable. #335876: Make the chat input workspace folder Remote picker easier to navigate with the mouse
-
Add the editor.wordWrapIndicator setting to show a marker at the wrap column for soft-wrapped lines. #47855: Add wordwrap indicator
-
Improve Auto model selection in multi-turn Copilot harness conversations with updated context routing. #333261: Support Multi-turn context routing v2 in Auto via Copilot Harness
-
Restore the Markdown editor option in the editor overflow menu for Markdown files. #334429: Markdown Editor is entirely unavailable
-
The proposed
AuthenticationSession.expiresAfter
API exposes an access token's lifetime in milliseconds, orundefined
when the expiry is unknown. #335184: API proposal: Expose AuthenticationSession access token lifetime
We really appreciate people trying our new features as soon as they are ready, so check back here often and learn what's new.
Facts Only
Visual Studio Code version 1.139 (Insiders) released on September 20, 2026.
Open Keyboard Shortcuts and Running Extensions as modal editors in the Agents window.
Rename Preview settings for completed agent sessions from Archive to Mark as Done.
Add /.jj to default files.exclude patterns to hide Jujutsu repository metadata.
Add model picker controls for choosing a model's Thinking Effort and Context Size in Copilot Chat.
Preserve an explicit Mark as Unread state for the focused agent session until read or left.
Remove stale persisted references and artifacts from agent sessions without deleting underlying files.
Support for Auto Tiers with Efficiency, Balance, and Intelligence options for model balancing in Copilot sessions.
Desktop runtime updated to Electron 43.6.0.
Added keyboard shortcut to mute or unmute the microphone while experimental Voice Mode is connected.
Added Cargo completions to terminal suggestions.
Executive Summary
Full Take
The development trajectory suggests a focus on refining developer workflows, particularly around AI agent interactions and system management within the IDE environment. The introduction of granular control over model behavior (Thinking Effort, Context Size, Auto Tiers) and session management (Mark as Done, stale reference removal) indicates an effort to move from simple functionality toward a more controlled, accountable experience for complex multi-step tasks facilitated by AI agents. This pattern suggests a systemic shift toward making the opaque nature of AI interaction more explicit and manageable for the user.
The integration of agent-specific controls that bypass standard session states, such as enabling agent merge directly on a PR, speaks to a desire to embed workflow logic directly into the interface layer. The technical updates, like updating Electron and adding specialized input controls for Voice Mode, demonstrate an ongoing commitment to platform stability while layering in highly specific, high-fidelity interactions. The tension here lies between enhancing the power of AI tools and ensuring cognitive sovereignty—how much control should the user have over the abstracted reasoning process? If the defaults are heavily tuned for efficiency (Auto Tiers), the risk exists that optimizing for speed might inadvertently obscure the nuance required for critical decision-making during complex agent operations.
What happens when efficiency settings inherently limit the visibility into reasoning, and how does maintaining an explicit "Mark as Unread" state reflect the value placed on session status versus operational speed? How are these fine-grained controls layered upon systems that manage external context and execution?
Sentinel — Human
The text appears to be a standard, fact-based release note compilation, likely human-authored documentation, even though it uses the mention of GitHub Copilot and AI tools in its framing.