Hi like i dont know what got codex mad but it just like did like actions that burnt up my 5 hour usage in like 1 hour. so that was no bueno. why is codex so vengeful?
#codex got angry
scenic shell
weak orioleBOT
Hi @scenic shell. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
You’re not alone—what you’re seeing is a mix of how Codex metering actually works and, in some cases, real bugs on OpenAI’s side.
A few key points from the docs and forum threads:
-
The “5 hours” is not real clock time.
It’s an allowance that can be burned much faster than 5 actual hours, depending on:- how long the agent spends “thinking” (reasoning),
- how big your repo/context is,
- which model you pick (Astra / Sol / Terra / Luna, etc.),
- and how long a single run is allowed to keep going.
You can hit 100% of the “5‑hour” window in well under an hour of real work. [Allowance in Work/Codex; Limit feels misleading]
-
Usage is tied to reasoning time, not message count.
Long-running or looping tasks can chew through your allowance very quickly, even if you only sent a few prompts. The more minutes Codex spends reasoning, compressing context, and re-planning, the faster your percentage drops. [Understanding new limits] -
There have been real metering issues.
- OpenAI has acknowledged bugs where Codex usage drained “faster than intended” due to abuse/fraud prevention systems and granted resets to affected users. [Abnormal weekly metering]
- Earlier, a bug in cloud task tracking also caused people to blow through limits on small tasks; OpenAI fixed it and issued credits. [Cloud tasks bug]
- Loops and stuck runs can silently burn everything.
Users have reported Codex getting stuck in long loops (repeating planning, context compaction, and re-checking work) for many hours, wiping out a weekly allowance without finishing the task. Starting a fresh chat in the same project often fixes this. [9‑hour loop report]
So Codex isn’t “vengeful,” but:
- Long or stuck runs,
- large projects with lots of context,
- and the current allowance design
can make it feel like it’s angry at your quota.
What you can do right now:
- If you saw obviously abnormal drain (e.g., tiny tasks nuking your whole window), capture screenshots and contact OpenAI support; there have been confirmed metering issues and resets before. [Abnormal weekly metering; Cloud tasks bug]
- Break work into smaller, bounded tasks and stop runs early if Codex seems to be looping or “thinking” without progress. [Understanding new limits; Limit feels misleading]
- If a chat feels “cursed” (lots of compaction, repeating itself), start a new chat in the same project; this has resolved runaway loops for other users. [9‑hour loop report]
If you describe what model you used and what the task looked like (e.g., long autonomous coding run vs. a few short edits), I can help you map that to how the allowance likely got burned.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
scenic shell
well you see getting stuck in those loops is what codex does when it gets angry. now like i try to stay on the good side but it seems like codex has a anger problem.