Agent and Codex usage appear to share a weekly quota without disclosure or high-consumption warnings #33690

Summary

Agent mode and Codex usage now appear to draw from the same weekly usage allowance, without a clear disclosure, a usable real-time meter, or a warning before a run consumes a substantial part of the remaining weekly quota.

On a ChatGPT Plus account, the weekly allowance reset on July 15, 2026 at approximately 15:23 CEST. After roughly two hours of mixed Codex and Agent usage, access was blocked at approximately 17:27 CEST until July 22, 2026 at 15:23 CEST.

Previously, Agent mode appeared to have its own separate limit: Agent could become unavailable while Codex remained usable. The current behavior therefore looks like either a newly shared pool or cross-product charging that is not made sufficiently visible to the user.

User impact

• A weekly allowance can be exhausted within a very short session immediately after reset.
• There is no meaningful warning that starting or continuing a task may consume most or all of the remaining weekly quota.
• The user cannot see how much quota is being charged separately to Agent mode, Codex, individual models, or individual runs.
• Long-running tasks may stop before producing a complete or usable result, while still consuming a disproportionate amount of the weekly allowance.
• A reset can therefore function like a one-shot event: the user resumes normal work, receives no warning, and is then locked out for nearly a week.

The problem is not merely that limits exist. The problem is that the accounting is unpredictable and the product provides too little information for a paying user to make an informed decision about whether a task is worth starting.

Expected behavior

At minimum, the product should:

1. Clearly disclose whether Agent mode and Codex share the same allowance.
2. Show remaining quota and reset time before a run starts.
3. Show which product, model, and run consumed the allowance.
4. Warn before a run is likely to consume a substantial portion of the remaining weekly quota.
5. Require explicit confirmation when a single run may consume most of the remaining allowance.
6. Avoid charging failed, aborted, or incomplete runs as though they delivered a complete result, or provide an appropriate quota adjustment.
7. Provide predictable per-run or per-period caps so one task cannot silently consume almost an entire weekly entitlement.

A warning could be as direct as:

This run may consume a large portion of your remaining weekly Codex allowance. Continue?

Environment

• Subscription: ChatGPT Plus
• Products involved: ChatGPT Agent mode and Codex
• Codex surfaces used: desktop/remote/CLI workflow
• Platform: macOS and iOS clients
• Observed: July 15, 2026

Related issues

There are existing reports about unusually fast or incorrect quota depletion, including #30212, #31174, and #28879. This report is specifically about the apparent cross-product pooling of Agent and Codex usage, the lack of disclosure, and the absence of a pre-run high-consumption warning.