[Bug] New Codex conversations fail with "Selected model is at capacity" while existing conversations remain usable on the same Pro account #45648

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

26.908.70816

What subscription do you have?

ChatGPT Pro, x20

What platform is your computer?

Windows 11, x64

What issue are you seeing?

Newly created Codex conversations consistently fail with:

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

However, existing conversations that had previously started successfully
continue to work normally on the exact same account, PC, network, and Codex
installation.

The affected Pro account still has approximately 45% of its weekly Codex
allowance remaining.

I have reproduced the capacity error across multiple available models,
including:

• GPT-6 Astra
• GPT-5.6 Sol
• Terra
• Luna

Switching models does not resolve the issue.

The behavior appears to be account-specific:

• Affected account + same PC/IP -> new conversations fail
• Different account + exact same PC/IP -> new conversations work normally

The most unusual behavior is:

• Existing conversation on affected account -> "say hello" -> works normally
• Newly created conversation on affected account -> "say hello" -> capacity error

I created multiple new conversations with only the trivial prompt
"say hello", and they consistently failed.

Therefore the failure does not appear to depend on prompt complexity,
repository context, or a specific model.

Signing out and switching accounts does not reliably resolve the problem.

What steps can reproduce the bug?

Steps to reproduce

1. Sign in to the affected ChatGPT Pro account in Codex Desktop.

2. Confirm that weekly Codex usage allowance remains available
   (approximately 45% remaining in my case).

3. Create a new conversation.

4. Select any available model.

5. Send the prompt:

   say hello

6. The new conversation fails with:

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

7. Open an existing conversation that had previously worked.

8. Send the same prompt:

   say hello

9. The existing conversation responds normally.

10. Sign in with another account on the same PC and IP address.

11. Create a new conversation.

12. The new conversation works normally on the other account.

What is the expected behavior?

Expected behavior

New conversations should be able to start normally while the account has
available usage quota.

If a specific model is temporarily at capacity, switching models should
allow the request to proceed.

Existing and newly created conversations on the same account should not
show different availability behavior without a clear explanation.

Actual behavior

Existing conversations remain usable, but newly created conversations on
the affected account consistently fail with the model-capacity error,
including trivial prompts and multiple different models.

Additional information

Additional information

This may be related to account-level admission/routing or new-conversation
provisioning rather than per-model capacity.

Related issues with overlapping symptoms:

• Codex Desktop unusable: "Selected model is at capacity" for all models on my account #39643 — account-specific capacity errors; other accounts work on same device
• Account-specific capacity errors across Codex models despite a fully available weekly allowance #43337 — account-specific capacity errors despite available weekly allowance
• Selected model is consistently at capacity on Codex App/CLI #43688 — capacity error across all models
• Bug Report: Unable to Create New Conversations in Codex Desktop (thread/start Timeout) #30348 — existing conversations work while new conversations fail, although
  that report shows a different thread/start timeout error

I have screenshots showing:

1. ChatGPT Pro weekly usage with ~45% remaining
2. Multiple newly created "say hello" conversations failing with the
   capacity error
3. An existing conversation on the same affected account successfully
   responding to "say hello"

The issue was observed on September 15, 2026 (KST, UTC+9).

OpenAI Support asked me to report this directly through the Codex GitHub
issue tracker for developer investigation.