Desktop app: add mobile-style response menu with model details and non-destructive retry history

Open 💬 0 comments Opened Jul 19, 2026 by abcjhby2871

What variant of Codex are you using?

Codex / ChatGPT Desktop App

What feature would you like to see?

Please bring the mobile app's per-response ... menu and response-version workflow to the desktop app.

The desktop app should support two closely related actions on each assistant response:

  1. Show the model used for that specific response
  • The response menu should display the actual model that generated the response, rather than only the model currently selected in the composer.
  • Where applicable, it would also be useful to show the reasoning-effort setting or execution mode used for that turn.
  • This metadata should remain attached to historical responses even if the conversation later switches models.
  1. Retry / regenerate without discarding the previous response
  • Retrying a response should create a sibling version instead of replacing or deleting the existing answer.
  • Users should be able to move between versions using previous/next controls or a compact version selector, similar to the mobile app's rolling response-version UI.
  • The selected version should persist after reopening the conversation.
  • It should be visually clear which response version is currently active.

Why this matters

In longer Chat and Codex workflows, users often need to:

  • verify which model produced a particular result;
  • compare two generations of the same answer;
  • retry a weak response without losing a useful earlier version;
  • switch models during a thread while retaining accurate per-message provenance;
  • return to an earlier response version when the retry is worse.

The current desktop workflow makes this harder because model selection is mostly conversation/composer-level, and retrying does not expose the same non-destructive version-navigation experience available on mobile.

Suggested interaction

From the ... menu on an assistant response:

  • Model details
  • Retry response
  • optionally Retry with another model

After retrying, show a compact control such as:

‹ 1 / 3 ›

so users can inspect every retained response version.

Codex-specific behavior

For agentic turns that ran tools or modified files, retry semantics should be explicit:

  • preserve all conversational response versions;
  • do not imply that previous filesystem or tool side effects were automatically reverted;
  • clearly indicate whether a retry creates a conversational branch, re-runs tools, or continues from the current workspace state;
  • retain the model metadata for each version independently.

Expected result

Desktop and mobile should provide a consistent per-response experience for inspecting model provenance, retrying an answer, and navigating retained response versions without destructive replacement.

Additional information

This request is specifically about message-level response metadata and version history, not merely exposing the current model picker at the conversation level.

View original on GitHub ↗