Codex main quota exhaustion blocks GPT-5.3-Codex-Spark despite available Spark quota

ChatGPT Support
1

Summary

When the main Codex quota is exhausted, GPT-5.3-Codex-Spark cannot be used even though its quota is still available.

The UI allows selecting GPT-5.3-Codex-Spark, but all requests are blocked with the message:
“You’re out of Codex messages”

Environment

  • Platform: Windows Desktop Codex App
  • Version: 26.609.41114 (Released June 13, 2026)
  • Plan: ChatGPT Pro
  • Projects: Reproducible in both existing and new projects

Steps to Reproduce

  1. Use GPT-5.5 or other Codex models until the main Codex quota is exhausted
  2. Confirm GPT-5.3-Codex-Spark quota is still available (e.g., 100%)
  3. Switch the model to GPT-5.3-Codex-Spark
  4. Attempt to send a message or run a task

Actual Result

  • The UI shows:
    • Model successfully switched to GPT-5.3-Codex-Spark
  • However:
    • All actions are blocked
    • Error message: “You’re out of Codex messages”
    • No execution is possible

Expected Result

GPT-5.3-Codex-Spark should remain usable independently of the main Codex quota.

If Spark quota is available, users should be able to continue working with it.

Additional Observations

  • Initially, GPT-5.3-Codex-Spark did not appear in the model list
  • After some time, it appeared, but still could not be used
  • The issue persists in newly created projects (not context-related)
  • ChatGPT (non-Codex) models like GPT-5.3 work normally

:backhand_index_pointing_right: Important:

  • In May 2026, this workflow worked correctly
  • After exhausting GPT-5.5, I was able to switch to GPT-5.3-Codex-Spark and continue working without issues
  • This suggests the current behavior is a regression or recent change

Suspected Cause

Possible quota handling issue:

  • The system appears to block all requests when main Codex quota = 0
  • Spark quota is ignored in the execution check

Or:

  • UI state and backend quota state are out of sync

Impact

  • Users cannot use available Spark quota
  • Quota display is misleading (Spark shows available but is unusable)
  • Breaks expected fallback workflow (5.5 → Spark)

Request

  • Ensure Spark quota is handled independently from main Codex quota
  • Fix quota validation logic so Spark remains usable when available
  • Investigate Windows Codex app state synchronization
2

To be honest i think this was a bug abuse fix. I never saw a separated quota for a dedicated model of codex anywhere on my end. I guess lucky you for having been able to use it for some time for free.

3

This is officially provided.

image
image708×396 6.43 KB

4

Where do you find that view ? In europe it’s only this view

image
image715×713 14.4 KB

5

To convert existing conversations to Spark, you must first compress the context into a generic model. However, switching to Spark may become impossible once the generic model limits are exhausted. In other words, you must have generic model capacity to switch to and use Spark models. It appears advisable to utilize Spark model limits first before using generic models..

image
image677×372 6.36 KB

6

Once again this option isn’t available in Europe or at least not for French Plus users. But having to compress it into a new chat or just to switch model sounds like a real pain.

7

The cause of this bug is that codex is using 5.4 to do the “approve for me”, and it is blocked when you are out of the normal quota. To fix this bug, they should configure 5.3-code-spark to do the approval. A temporary alternative is to toggle “approve for me” to “ask for approval”. But this is still a real bug that needs fixing and highly hinder user experience.