Did any one notice way higher than expected token consumption report in api usage with gpt 5.5 today (Apr 28).
My dashboard suggests that when using codex I used 95 Million tokens input, 25 Million cached input to produce 600K output tokens!!??
I did not use codex for such intense work, and the output token count to input ratio should on its own suggest there is some kind of problem with this.
Hey @kurlytail, yeah, that spike would feel pretty off, especially with that input to output ratio.
There’s actually been some discussion around usage reporting, you might find this thread helpful: Codex Rate Limits Discussion Thread. A few folks are comparing notes there and sharing what they’re seeing.
Worth a quick skim to see if it lines up with your numbers, and to track any updates as more info comes in.
-Mark G.
Codex 5.5 is consuming tokens as if its drinking gasoline even at thinking medium
Openai should seriously look into this problem
Either come up with a codex version or decrease the token consumption
Hey guys, you have to fix the issue ASAP. Don’t play the same games as Anthropic, please.
The most I was able to do before was to hit my weekly limit after 6 days which felt pretty reasonable to me. Now with using GPT 5.5 (they even buried 5.3 Codex in another submenu drop down together with touting using fewer tokens with 5.5 so i did switch to using 5.5) I managed to hit my weekly limit within 2.5 days. That does not feel great guys.
I’ve also experienced this problem .Our token usage has gone up 5 fold with 5.5.
Hi. Guys this should be fixed asap. I used it for like 5 mins and its gone. I was in gpt-5.4 high fast
It just burns the token so much faster since the beggining of the week. Last week it was still fine. But this week it really run dry extremely fast. And the “5h limit’” is disappearing in 30 mins.
Yes, I agree with you. Today I also submitted a complaint because I believe Codex is misleading its users and basically robbing them. The interface says “5 hours of work,” but in real use with a serious project, that limit can be burned through in about an hour. So “5 hours per month” does not feel like honest working time. It feels like a very limited usage budget that disappears much faster than users expect. I am extremely dissatisfied with this. On top of that, when Codex starts, it reads AGENTS.md and possibly other system or instruction files, and from what I understand, this already consumes part of the limit. In my case, around 3% was gone before I even started doing any real task. If Codex does this on every launch or every new prompt/task, then the limits burn even faster. This means users lose part of their paid quota simply because Codex reads setup files and prepares context, not because actual work has been done. I consider this robbery. This should either be clearly explained in the interface, or these internal setup costs should not be counted against the user’s limit.
its true. i am feeling the same thing.
There seems to have been an issue with Codex 26.513.31313
Looks like they patched it in 26.513.40821
Unfortunately its seems to have consumed tokens and highly accelerated rate. Is there plan to do a reset?
This is a serious issue, I give up from Claude code, but codex now playing the same thing. The limi is just a trap.
Where do you see which version of codex is running? And where did you get the detail patch note of the versions ?
It’s amazing how since 5.2 the Token Usage skyrocketed.
And at this very moment, the 5.5 version is like Claude…
So, i’m not sure what games are we playing, but this is abnormally low, after 2 commands, simple ones for a script edit, i end up with 51% left, working locally, VSCode, Extension Agent.
This made me come here and shout a bit, because i’m not the type that usually comes, but it is so bad that it is needed to complain about it !
Please have a look at token usage, the change from 5.2 to 5.5 is dramatic.
The change from 5.2 to 5.3 is bad as well
Change from 5.2 to 5.4 was more acceptable, but still high, although lower than 5.3
But on 5.5 … we are equaling Claude, which to be honest, i closed it in day 1 after 3 commands and running out of tokens 
@Marius91 The issue isn’t the tokens — it’s the plan itself. People on the forum basically explained to me that the lower-tier plan is only enough to “warm up” Codex, not to seriously work with code.As I said before, I want a clear explanation of how tokens are spent. But apparently, that is also considered my problem: I’m supposed to dig through system files myself to understand why the tokens disappeared so quickly and why I hit the limit.And if everything depends so much on saving tokens, then Codex should also help users save them: optimize prompts, avoid unnecessary output, keep responses compact, and clearly show what consumes the limit. Because right now users are forced to economize on every request and every answer just to keep working. 
