After Codex went down on September 26, Work was restored, so I sent a single message just to test it. 6Astra replied with only a few lines, but that short exchange alone consumed 20% of my 5-hour usage limit and 5% of my weekly usage limit.
I thought maybe the usage had been deducted during the outage, but then I created a new topic in Work and sent another single message. Again, with only a response of a few lines, my 5-hour usage limit decreased by more than 30% and my weekly usage limit decreased by 9%.
Is anyone else experiencing the same issue? At this rate, I cannot use Work for my actual work.
Hey hirohamada, that's indeed frustrating.
Could you confirm your plan and whether the usage you're referring to is included with your plan or comes from separately purchased credits?
Avinash
I’m on the Plus plan. It’s the 5-hour and weekly usage limits that come with the plan.
After Codex recovered from the outage on September 26, just two exchanges — a single short message from me and a few lines back from the model — caused nearly 20% of my weekly usage to drop all at once.
It seems to have gone back to a normal consumption rate for now, but I’m afraid this could happen again at any time, so I keep checking my usage over and over as I use it.
I’m experiencing the same issue currently.
I’m on the ChatGPT Pro $100 plan (Pro 5x), and the usage I’m referring to is the included weekly usage limit, and my 765 purchased credits.
Since the September 26 Codex outage/recovery, I’ve noticed my Codex usage dropping much faster than expected for the amount of work being performed. Normal or relatively short sessions appear to be consuming a disproportionately large percentage of my usage allowance.
This does not seem consistent with my previous Codex usage patterns before the outage.
I’m monitoring my usage percentages before and after sessions now so I can document the issue more precisely.
If OpenAI Support needs additional information such as the model used, reasoning level, timestamps, before/after usage percentages, screenshots, or the Codex environment I was using, I can provide that as well.
I wanted to add my report here because this appears to be affecting more than one user.
That's frustrating.
Could you share the session IDs for the affected sessions, along with the approximate timestamps and models used? That would help narrow down the issue.
Avinash