Major changes:

• First class support for DeepSeek V4 models
• Sections replaced with sub-projects, which makes context segregation in monorepos more manageable
  • Sections had no consistent adoption pattern and lacked analogues in Claude Code / Codex

Pleasantries:

• Sped up SSH runtime setup significantly
• Solved a variety of layout flashes and performance issues in transcript streaming

Removals:

• Removed the Orchestrator built-in agent
• Removed the file explorer

This release brings GPT 5.5 and GPT-5.5-Pro, supporting both standard API billing and Codex authentication.

This release brings Opus 4.7, which has the same UX as 4.6 with the exception of a new "xhigh" thinking in between "high" and "max".

Features

• Archive Improvements — snapshot-based worktree cleanup on archive, configurable Coder workspace action (keep/stop/delete), and a destructive confirmation for untracked-file failures
• Left Sidebar Redesign — tighter indentation, aligned status dots + sub-agent connectors, hover-only project row actions, subordinate workspace counts

Performance

• Optimized Immersive Review performance on large diffs
• Optimized SSH workspace creation on slow local connections
  • Replaced git bundle with native git push + delta negotiation
  • Sharded OpenSSH exec paths for parallel sync
  • Prefetches origin on SSH host before local→remote push
• Sorted tool definitions deterministically for prompt cache stability
• Avoided extra Anthropic cache breakpoints with explicit TTL

Fixes

• Fixed multiple cases where the chat transcript would vertically tear on chat switches
• Preserves all init logs across workspace switches
• Restores last page and open chat on hard refresh
• Restored Cmd+. agent cycling on macOS
• Reattaches loaded skills after compaction
• Fixed attach_file on macOS (unpacked sharp runtime assets) and added raster image scaling
• Restored archived workspaces with persisted branch mappings
• Stripped leading/trailing newlines from quote input text

Cleanup

• Removed the built-in "Chat with Mux"
  • Our goal is to replicate the capabilities of this chat in regular chat sessions
• Removed Auto mode

Browser & desktop workflows

Enabled in Experiments -> Agent Browser

• New Browser tab support for live agent-browser sessions, including an address bar, reload control, ready-state display, and streaming updates.
• Better Browser tab signal with clearer Recent Actions and less noise.
• Improved reliability for keyboard delivery, shell-launched browser sessions, and server-mode browser sessions.
• Added workspace-scoped virtual desktops via Portable Desktop.

Agents, SSH, and models

• Added grouped task variants and made task variants easier to discover.
• agent-browser is now available as a built-in skill.
• SSH workspaces can now use host-global skills and agents.
• Added support for GPT-5.4 Mini and GPT-5.4 Nano.

Quality of life

• Consistent recovery of interrupted chats on restart.
• Improved queued-message handling and chat layout stability while sending or resizing the composer.
• Polished immersive review with LoC progress, cross-file navigation, and undo.
• Reduced idle compaction / notification noise and improved startup diagnostics.

Sub-agents

• New sidebar hierarchy for sub-agents, including connector lines and better expansion for completed runs.
• Active sub-agents animate in the sidebar.
• The sub-agent toggle moved to the kebab menu.
• Double-click now always renames.
• Added support for "Best of N" type sub-agent launches where the parent launch N sub-agents with the same prompt and synthesizes the best path forward

Analytics

• Added a save action for chat analytics results.
• Added a saved analytics panel SQL editor.

Quality of Life

• Better handling of edited messages before stream start.
• Reduced false interruptions by not treating slow stream starts as interruptions.
• Todos are session-scoped now, so they survive stream interruptions better.
  • Completed todo lists are cleared more reliably after final streams.
  • Pinned todo panels now auto-collapse more appropriately.
• Better wide-touch sidebar interactions.
• Mux now surfaces MCP server startup failures instead of failing more silently.
• Fixed some cost display/reset issues around Codex OAuth and forked workspaces.

What's Changed

• improved cost and tracking management for gpt-5.4 family

Full Changelog: https://github.com/coder/mux/compare/v0.20.0...v0.20.1

✨ Highlights

• API support for gpt-5.3-codex and added support for gpt-5.3-codex-spark
  • Note: at the time of this release Spark requires the ChatGPT Pro ($200/m) plan
• Analytics page (Cmd+Shift+Y) to provide a convenient look into spend and model performance
• "Auto" agent mode which delegates selection of exec/plan/ask/etc. mode to the agent
• Immersive Review (Shift+I), which provides an IDE-like, full-file review experience
  • We are actively fixing some performance issues with large files, consider this feature a preview of things to come
• Nightly release channel available when clicking the version number in the top left of Mux

Misc.

• Contains many improvements to the mobile experience
• Simplified the Chat Input controls, consolidating help indicators
• Added first-class support for creating creating skills in the Chat with Mux help workspace

Another polish/fix-oriented release with some notable changes:

• First-class support for Sonnet 4.6 and Gemini 3.1 Pro
  • Sonnet 4.6 has much of the same new features and capabilities as Opus 4.6, including 1M context, max thinking, and effort-based reasoning
  • Available via aliases sonnet and gemini
• Text now streams in more smoothly
• Many improvements to the Orchestrator agent
• Fork improvements
  • Now well-supported for Docker workspaces
  • Available as a quick action in the workspace context menu
  • Removed the name requirement, so /fork [message] works and auto-generates a name
• Title improvements
  • Hit F2 to quickly re-title a chat
  • Hit Shift+F2 to auto-generate a new title
• You can now right-click text in the chat transcript and then click "Quote in input" to quickly address a portion of the response
• Support for dragging and dropping PDFs as input for Opus 4.6

GPT-5.3-Codex

We’re releasing this off-cycle to make GPT-5.3-Codex broadly available to users. Support is enabled via our new ChatGPT plan settings at Settings → Providers → OpenAI.

As of now, GPT-5.3-Codex ranks #3 on Terminal Bench with a strong showing on the Terminal Bench leaderboard.

Chat Sharing

You can now share a full transcript of your chat via an end-to-end encrypted link.
Right-click the workspace and select “Share transcript.”

Here’s a sample chat.

Misc.

This release includes several additional improvements:

• Workspace chats are now append-only on disk, preserving history prior to compaction.
• In this release, you can view up to one compaction back; full paging support is coming soon (including mux.md sharing).
• The SSH runtime now uses remote worktrees, making new workspaces and forks much faster.
• Skill load errors now appear directly in the workspace’s skill badge.
• mux run now supports sub-agents.
  • Invoke via bun x mux run or npx mux run.

Key changes:

• opus alias updated to the just-released Opus 4.6
  • Including support for 1m context, adaptive reasoning, and max thinking level
• You can now one-shot messages with a particular model, e.g. /haiku update PR body to quickly/cheaply execute a task
• Cleaned up the Chat Input controls, with much better presentation on mobile viewports
• Many bug-fixes, QoL improvements and UI polish

🏙️ Highlights

Left sidebar has been simplified, removing the built-in Runtime badges and offering way more information on hover:

Unfinished workspace prompts now become draft workspaces in the sidebar:

Remaining gateway credits now visible from the Chat Input:

We now have a built-in help chat workspace for any questions about Mux setup or behavior:

Context usage breakdown by file now available in Costs tab:

🛠️ Other stuff

• Improved the mobile view
• Improved performance on busy sessions via more renderer heap (da1c049f3)
• Made compaction more reliable (f75a0cf40)
• Name generation is more robust, correctly respecting gateway prefs (a13decca4)
• Fixed bug where chat-level compaction slider would disappear right as the mouse reached the slider 😅 (cd010556c)

🌃 Highlights

You can now use slash commands with skills.

For example, if you have a skill called pull-requests, you can run:

/pull-requests open PR

Mux will automatically load that skill into the prompt before sending your message:

We now warn when the auto-compact threshold is higher than the model’s output reserve, helping you avoid “context exceeded” errors:

Additionally, if compaction fails on GPT-5.2 or Sonnet 4.5, we automatically retry once with more permissive options:

• GPT-5.2: Auto Truncate enabled
• Sonnet 4.5: 1M context enabled

No configuration required—this should work well out of the box.

We made several improvements to the review flow, including:

• Highlighted lines stay highlighted when the review is staged
• You can add review comments from the FileViewer (not just diffs)

Misc.

• Minor improvements to the mobile layout (more coming soon)
• Fixed a thinking-level persistence bug when using XHIGH
• New /debug-llm-request command to show the full, unmodified payload of the most recent model request for debugging and analysis

Highlights

• PR status and link now available in real-time in the workspace chat view:
  
  (requires authenticated gh CLI)

• Added file explorer and minimal file viewer:
  

• We now support a first-class Coder Workspaces runtime (docs)

• Built-in support for Codex 5.2 now available (/model codex)

• Fixed a variety of Windows bugs

Features

• Login with Mux Gateway is now live!
  If you have a coupon code you can still redeem them at https://gateway.mux.coder.com/
  When launching mux you can login to the gateway to automatically use the vouchers.
  

You can also join our discord to talk mux and request some tokens for testing!

Fixes

• improvements to review pane - file status, CRLF handling and inline review comments
• revamp of the provider menu for easier management of custom models
  

Full Changelog: https://github.com/coder/mux/compare/v0.12.0...v0.12.1

🏙️ Highlights

We now support integrated terminals in the right sidebar, with drag-and-drop tiling and resizing:

Terminals persist on the backend, so mux server users can leave a tab, come back later, and resume the session exactly where they left off.

New runtime: Docker!

mux run has been given a makeover to make it useful for automation and CI. See:

• mux run docs
• Example GitHub Actions workflow

🛠️ Fixes

• Unified handling of environment variables (docs)
• Better default $PATH on macOS using common shell profiles (49176ed39)

🌃 Highlights

Project now support customizable sections for organization:

You can share plans and other assistant outputs from Mux via the e2e encrypted mux.md:

Idle-compaction settings moved next to usage-based auto-compact slider:

When reaching a context exceeded error, we now give you a more useful retry operation if you have a model available with a higher context limit:

See also:

• Fully customizable Agents (docs)
• Mux can notify via new notify tool, just instruct via prompt (docs)
• Tool hooks! (docs)
• You can tag files with "@" in the chat which will include them in your context

🛠️ Fixes

• Fixed bug where SSH workspace sub-agents had unexpected git state (b4125da56)
• Fixed thundering herd issue where enabling idle compaction could lead to dozens of old workspaces undergoing simultaneous compaction
  • New behavior: only one idle compaction runs at a time (f24db303f)
• Many performance improvements all around the app