Multiple GPT-5 / GPT-6 models return "Selected model is at capacity"

Description

Codex repeatedly returns the following error across multiple models:

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

This does not appear to be limited to a single model. I observed the issue when switching between multiple GPT-5 and GPT-6 models, including GPT-6 Astra and GPT-5.6 variants.

Actual behavior

Requests fail with:

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

Switching between available GPT-5 / GPT-6 models often results in the same error.

Retrying does not reliably recover the request.

Expected behavior

If one model is temporarily unavailable, Codex should route the request to another available backend or provide more detailed information about the failure source.

It would be helpful to distinguish between:

• individual model capacity
• account-level throttling
• regional capacity
• backend serving/routing failures

Reproduction

1. Open Codex.
2. Select GPT-6 Astra or GPT-5.6.
3. Submit a normal coding task.
4. Receive Selected model is at capacity.
5. Switch to another GPT-5/GPT-6 model.
6. The same error can occur again.

Additional information

The problem affects multiple models at the same time, suggesting this may be related to Codex backend capacity or request routing rather than one specific model being unavailable.

A request ID or more detailed diagnostic information in the UI would help users report these incidents.

Date observed

2026-09-07

The report names no files, tests, or entry points. Start by searching for the exact capacity error and tracing model selection and request routing; done means identifying whether the failure is model, account, regional, or backend-related and documenting or testing the resulting behavior.