[Windows][Codex Desktop][Pro] All models fail with "Selected model is at capacity" (server_overloaded)

What version of the Codex App are you using (From “About Codex” dialog)?

26.903.9818.0

What subscription do you have?

ChatGPT Pro 20x

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What issue are you seeing?

Codex Desktop is currently unusable on my ChatGPT Pro account because all available models repeatedly fail with:

Selected model is at capacity. Please try a different model.

Switching models, creating new sessions, and retrying do not resolve the issue.

I inspected the local rollout JSONL files generated by Codex Desktop. The underlying error is consistently:

"codex_error_info":"server_overloaded"

Example:

"error": {
  "message": "Selected model is at capacity. Please try a different model.",
  "codex_error_info": "server_overloaded"
}

### What steps can reproduce the bug?

1. Launch Codex Desktop on Windows.
2. Sign in with a ChatGPT Pro account using ChatGPT authentication (not an API key).
3. Start a new session.
4. Select any available model.
5. Submit a minimal prompt, for example:

   `Reply with exactly: OK`

6. The request fails with:

   `Selected model is at capacity. Please try a different model.`

7. Create another new session and/or switch to another available model.
8. Submit the same minimal prompt again.
9. The same error occurs across multiple models.

The issue is also reproducible in fresh sessions with very little context, so it does not appear to depend on context-window usage.

Inspecting the corresponding local rollout JSONL files shows the underlying error:

`"codex_error_info":"server_overloaded"`

Example:

```json
"error": {
  "message": "Selected model is at capacity. Please try a different model.",
  "codex_error_info": "server_overloaded"
}

### What is the expected behavior?

Codex Desktop should successfully process requests when my ChatGPT Pro account has available usage and the selected model is available.

If the backend is temporarily overloaded, Codex should either retry successfully or return a clear server-overload error.

The same `server_overloaded` failure should not persist across all available models and newly created sessions for an extended period of time.

### Additional information

<img width="1149" height="114" alt="Image" src="https://github.com/user-attachments/assets/e4dcb2d6-019c-4e40-8b58-6eb2da74a803" />

Start by reproducing the failure in Codex Desktop on Windows with a ChatGPT Pro login, then inspect the local rollout JSONL files mentioned in the report. The issue names no repository file, test, or client-side entry point, so first determine whether the failure is actionable in this repository; done is a confirmed fix or a clear backend ownership boundary.