openai / openai/codex

[Desktop][Launch blocker] Stale model_catalog_json silently hides newly launched models

Open
#43,052 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

CLI config enhancement
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

Impact: new model launches are effectively broken in Desktop

For Desktop users with model_catalog_json configured, a newly launched OpenAI model can remain completely absent from the model picker even after updating and restarting Codex.

The app otherwise appears healthy and provides no error, warning, or indication that a local startup catalog is overriding the current bundled/server catalog. From the user's perspective, the new-model rollout simply never arrives. This breaks the primary Desktop flow for discovering and selecting newly launched models and is easily misdiagnosed as an account entitlement or staged-rollout problem.

This failure mode can recur at every new model launch until the user manually discovers and removes or updates the stale catalog override.

Affected variants

  • Codex Desktop app on Windows
  • Codex CLI on Windows and Linux
  • Reproduced with Codex 0.153.4 and the GPT-6 Astra launch

Reproduction

  1. Configure model_catalog_json with a valid custom catalog containing the currently available models.
  2. Later update Codex to a version whose bundled/server catalog includes a newly launched official model.
  3. Open the Desktop model picker or CLI /model selector.
  4. The newly launched model is missing indefinitely, despite the current client and account supporting it.
  5. Restarting or updating Codex does not help because the stale startup override continues to win silently.
  6. Remove model_catalog_json and fully restart Desktop/CLI.
  7. The new model immediately appears.

Confirmed example

With Codex Desktop 0.153.4:

  • the bundled catalog contained gpt-6-astra with list visibility;
  • the signed-in account could successfully run Astra;
  • a custom catalog created before the launch did not contain Astra;
  • Desktop silently used the custom catalog and omitted Astra from its model picker;
  • removing the override and restarting immediately restored Astra.

This confirms that neither client support nor account entitlement was the cause.

Expected behavior

Desktop should not silently look current while an old startup catalog hides newly launched official models. At minimum, it should:

  • prominently mark the model picker as using a custom catalog;
  • compare the custom catalog with current bundled/server models;
  • show a launch-visible warning such as: “Your custom model catalog is hiding 1 newly available OpenAI model.”;
  • identify the catalog path and modification time;
  • offer a Use current catalog or Review override action;
  • surface the same diagnosis in /status and codex doctor.

The custom catalog can remain authoritative; the important requirement is that newly supported official models must not disappear silently during launches.

Related issue

#20492 tracks stale remote app-server versions after upgrades. That can amplify launch problems, but this report specifically covers Desktop and CLI silently hiding new models because of model_catalog_json precedence.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the stale model_catalog_json behavior in Codex Desktop and the CLI, including the model picker, /model, /status, and codex doctor surfaces mentioned in the report. Done means users are clearly told when an override hides newly available official models, with the catalog path and modification time shown and an action to use or review the current catalog.

Written by the indexing model from the issue text.

Assessment

Tech stack
rust
Domain
cli, desktop
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.