I have an active ChatGPT Pro subscription. My General weekly allowance is exhausted, but GPT-5.3-Codex-Spark still shows 100% of its 5-hour allowance and 75% of its weekly allowance.
I explicitly selected “GPT-5.3 Codex Spark High,” but the app displays “You’re out of Codex and Work usage.” The error points to September 19 at 3:24 PM, which matches my General reset—not Spark’s weekly reset at 10:06 PM.
Please investigate whether the General usage check is incorrectly blocking Spark, or explain the restriction that applies. Please also confirm whether the displayed remaining Spark allowance is accurate.
This is a useful meter/gating specimen because the UI is apparently exposing two different allowance states at the same time: General exhausted, while Spark still shows remaining 5-hour and weekly allowance.
I would avoid assuming yet that those displayed Spark balances are contractually independent of the General gate, because that relationship is exactly what needs classification. The strongest next evidence is much smaller than another depletion test.
If you already preserved it, please keep/post a redacted screenshot showing:
the General allowance + reset;
the Spark 5-hour/weekly allowance + reset;
the exact “You’re out of Codex and Work usage” message;
the Codex surface (desktop / CLI / IDE) and app/version;
the selected model/effort.
A session/task ID is useful privately for Support, but there is no need to publish account IDs or anything sensitive.
The engineering/product question is simply: is remaining Spark allowance supposed to be usable after General weekly exhaustion, and if not, why is it displayed as independently remaining?
No need to deliberately burn another allowance to reproduce this. The current state is already the interesting part.
Yeah, it’s possible that when the main usage is depleted, it doesn’t let you use it at all, including spark. Try to preserve evidence and afterwards a support ticket…
I used yesterday 5.3 with no issue but now it is not longer in the model selector. Instead, using thesame thread where I used it yesterday it says that 5.3 is not supported when using a ChatGPT account.
In the model selector it says “Personalized” instead the actual model selected 5.3. Surprisingly the reasoning level it still says Alto (High in English).
At any other new thread 5.3 is not longer an option
