Zed editor gutter showing run-status indicators after test runs with pass and fail icons Image: Zed / github.com/zed-industries
by VibecodedThis

Zed 1.13.1: Agent Panel Response Controls, Mistral Thinking, and Dev Container Interop

Zed's latest stable release ships per-response buttons in the Agent Panel, thinking support for Mistral Medium 3.5 and Small 4, run-status indicators in the editor gutter, improved branch picker filtering, and fixes for a Git timeout bug and a Linux memory leak.

Share

Zed 1.13.1 landed on July 29 as the new stable release. The headline items are per-response action buttons in the Agent Panel and thinking support for Mistral models, but there are also a lot of quality-of-life fixes scattered through the release, including two that had been causing real friction: a Git operation timeout on slow hooks, and a memory leak on Linux with Fcitx5.

Agent Panel gets response controls

Every response in the Agent Panel now has three buttons: “Copy this Agent response,” “Scroll to User Message,” and “Scroll to top.” The scroll-to-user-message one fixes a bug where clicking it always jumped to the most recent prompt instead of the one attached to the response you’re looking at. More useful in long multi-turn sessions.

The panel also handles performance better with many file changes in a thread. If you’ve run agents across large diffs, the panel was getting sluggish.

Mistral thinking support

Mistral Medium gets bumped to version 3.5, and thinking mode now works for both Mistral Medium 3.5 and Mistral Small 4. The provider also drops three deprecated models: magistral-medium-latest, open-mistral-nemo, and devstral-medium-latest.

A separate fix resolves custom Mistral configurations rejecting Ministral model IDs, which was tripping up people who’d set up their own endpoints.

Run-status indicators in the gutter

Runnable gutter controls, the play buttons that appear next to test functions and runnable code blocks, now show the result of the last run. A check icon for pass, an error icon for fail or cancel, and an accent play icon while something is running. There’s also a “Clear Run Status” option in the context menu if you want a clean slate.

The same status update happens whether you run from the gutter button or from an inline code lens action like “Run Test.”

Branch picker improvements

Branch picker filtering, grouping, and branch creation suggestions all got work this release. Remote provider icons now show up alongside branches, which helps when you’re working with multiple remotes. This was a Zed Guild contribution from chirivelli.

Dev Container interoperability

Containers created in Zed can now be picked up by VS Code and the Dev Container CLI. Docker Compose dev containers also fix a bug where resuming an existing environment didn’t start all the configured services.

This lands from a large PR that was open for a while, and it’s a meaningful step toward Zed’s dev containers being interoperable rather than isolated from the broader toolchain.

GPT-5.6 Luna in ChatGPT subscriptions

Zed’s ChatGPT subscription provider now supports GPT-5.6 Luna, the most efficient tier in OpenAI’s latest model family. Luna sits at $1/$6 per million input/output tokens, making it cheaper to run than the full Sol tier while still using the same underlying 5.6 architecture.

Notable fixes

  • Git timeout: Git operations were failing with a misleading “Connecting to host timed out” error after 17 seconds. This affected large fetches, slow hooks, and agent-based authentication flows that prompt Touch ID. The 17-second cap is now gone.
  • Fcitx5 memory leak: Fixed excessive memory usage when composing text with Fcitx5 on KDE Wayland. A meaningful fix for Linux users on that stack.
  • macOS clipboard crash: A crash when reading from the clipboard on macOS.
  • Hunk staging corruption: Fixed a bug where staging a hunk could mark a different hunk as staged, and then unstaging could corrupt the Git index, when the diff had repeated lines.
  • MCP server directory: Agent MCP servers were starting from the wrong directory when launched before a worktree was available.
  • Duplicate GPT-5.x responses on Bedrock: Fixed colliding tool calls and duplicated replies when using GPT-5.x models on Amazon Bedrock’s Mantle endpoint.

Fast Mode deprecations

Fast Mode for Opus 4.6 and Opus 4.7 is deprecated in this release. With Opus 5 now the current model, the older fast mode variants are being wound down.


The full release notes are on GitHub. Zed 1.14.1-pre is also out in preview today, shipping sandboxing for agent terminal commands and web fetches. That one should hit stable in about a week.

Share