I also just started my work day after a very successful session yesterday on my $100 plan. Yesterday I could get a lot of work done in my five-hour windows. Today I hit my five-hour limit within one hour of question and answer and a few reviews and fixes. So no real feature development or anything, and still my limit is depleted. Also, my weekly is already down 40% within one hour. This can’t be right. This makes the investment totally unusable for real engineering work as a solo developer.
Hi Eric,
Sadly fixes can be the most demanding even when you think it is really easy fixing at first. It’s hard to know if it is a bug or normal with so few context.
Not saying it was not a bug. Just saying it’s hard to conclude if it was a real one or not.
Don’t get me wrong. I’m one of the front pusher for the 5h limit. But than you’ll have new commers that come and burn 1 week of quota in one wrong prompt and are locked out for a week.
I’m a plus user. But in my honest opinion, the 5h limit shall not exist for pro accounts that are supposed to know what they are doing.
Yeah I know how demanding my multigagent review heavy workflow is 
But it is just too obvious. Yesterday I did about 5 full multiagent review sessions plus fixes in my 5 hour limit and 2 parallel grilling sessions with a lot of context and today just one review, a few questions and a few fixes. Just after receiving my second banked reset it is like that…
It’s not a bug,
OpenAI has significantly reduced the Codex usage limits across almost all plan tiers; the new limits amount to only about 30% of the levels in place prior to June 18.
Again ? Do you have a public article or a source to that ?
Some people showed me the quota data they retrieved from sub2api, and quite a few people on Reddit are talking about it too.
Thanx for the info. I will use with this settings and I hope it reduces limit usage.
Assuming this is true, it was just done by about the same not even two months ago. Are you sure your not refering to the April changes?
While I totally agree with the whole thread, for sure something has changed again around the 16th June.
No, they reduced the limit three times. Before the end of the double limit in March, the weekly limit was reduced a bit, with that double limit was finished, the limits are reduced from $400-$500 to over $200. In April, they reduced it to around $180 by resetting the limit. The latest reduction was after the 618 reset, when many accounts had their weekly limit reduced to below $100. Furthermore, accounts under the same plan may have different current limits.
These examples are based on the credit limit of the Plus account. The credit limit for each account may have been different a long time ago, but the difference is obviously not as significant as it is now. Although the two tiers of the Pro plan do not strictly follow the nominal data, they are still proportional to the credit limit under the Plus plan (there may be a coefficient that we are unaware of affecting the credit limit obtained by each account, but I am not sure).
Thanks for the insight.
The April quota issues were very noticeable, and the problems I have seen since around 16 June feel just as bad, if not worse. Personally, I have been testing alternative providers since April because of the impact the earlier limits had on day-to-day coding workflows.
One temporary workaround was moving some users onto Pro with the higher usage allowance rather than relying on business accounts, but even then, people are still running into quota issues. That makes it difficult to rely on Codex-style tooling for sustained development work.
At this point, I think it is sensible for people to evaluate alternatives as part of their workflow. I do not necessarily mean the usual “big three” providers either. In my own testing, GLM 5.2 has been very capable for many Codex-like coding tasks, and the quota on their GLM Coding Pro tier feels much more reasonable for regular use.
The main caveat is data handling. I would not assume it is suitable for sensitive or proprietary code without carefully checking the terms and privacy position first. For less sensitive tasks, though, it can be a useful way to preserve limited Codex quota for the cases where OpenAI’s models are genuinely needed.
One other thing to note: it is primarily text/code-focused, so it is useful for development work, but less suitable for design-heavy or multimodal tasks.
I found the culprit: cache rate.
Check your cache rate carefully, you should have >90% during a long conversation, overall > 85% or 90%, otherwise your limit would be consumed at an incredible speed.
Same here.
I’m a Pro subscriber (20x), and I’ve used about 50% of my weekly quota in just one day. That’s far beyond anything I’ve seen previously, and my usage patterns haven’t changed enough to justify such a dramatic increase.
I’m honestly pretty concerned. I really hope this is a bug or a reporting issue rather than a permanent change to the quota system.
The frustrating part is that nobody seems to know what’s going on. People are reporting wildly different usage behavior, but there hasn’t been much official communication clarifying whether this is expected, a bug, or a policy change.
If something changed, just tell us. Most developers can adapt to new limits. What makes planning difficult is the uncertainty.
How to do that? I did not found a option for that in the codex app
Wow indeed sounds like you got redirected for the same price 
try ccswitch, you could find it on github
Because I don’t remember in which file it was.
Hi there!
I started using Codex this week with a Plus plan to evaluate it. Today, with just three simple code-checking tasks, involving fewer than 10,000 lines, I hit the limit three times and have only 2% of my weekly quota left. Are you experiencing this kind of abnormal usage?
I normally work with Claude, and even though its consumption is high, it doesn’t come anywhere near Codex’s usage levels. With Claude the same tasks took me 6% of my weekly…
Edit:
Just checked the usage, and today I used just 1.4M Tokens. Yesterday 14.8M with no issues. Two days ago 48M hitting the limit. Always using 5.5 at mid or high, never using fast option.
i have same error
pro , 200$
but only 5 minute work
