I’m a bit confused about how the limits are being applied.
This week, I never hit the 5-hour limit, but my weekly limit is already at 0%, even though there are still three days left before reset.
I ended up switching to Claude and Copilot so I could keep working. This has never happened to me before, so I wanted to check whether this is normal or whether something may have changed recently.
Is anyone else seeing their weekly limit get exhausted much earlier than expected?
I havent experienced this myself. Can anyone else shed some light on this situation?
5h limit and weekly limits are separate throttles
if you were already at 99% when you started into the 5h window, it’s normal that you’d hit the weekly limit well before hitting the 5h limit.
I recently bought a Plus subscription, but after just two or three requests, my 5-hour limit dropped to zero, and my weekly limit went down to 70. At the same time, Codex only executed two commands and spent the rest of the time analyzing files.
I was using the medium 5.4 model with a 1 million token context, and I’m confused about how the limits are calculated. If I wait 5 hours, will the same thing happen again—where just a couple of commands immediately exhaust the entire 5-hour limit?
Could you please clarify how this works?
So can anyone explain to me how it works? 15-20m he reads files and make one change and after this a see that i hit 5h limit to 0% and weekly from 100 to 70.
I dont have a answer to your problem
but i can walk you through a solution to extend codex operations and lessen the usage on context and rehydration. that way you can get alot more done. I think each one of my codex calls uses a few million tokens worth of data, but only uses codex for a few thousand tokens worth.
pretty confident in that. Granted I clearly burn through usage very fast. Think im close to 2 billion tokens a day in the stack.
you can make a custom GPT and export the data it gives you via connectors/MCP locally and then use that same bus to inject data back into the custom gpt, before having codex review it. LMK if you want the step by steps. been doing it for a year.
I appreciate the suggestion on how to improve this. However, the issue is that I didn’t experience this before — as recently as last week, I often couldn’t even use up the full 5-hour limit and could just keep working without hitting it. The weekly limit would last me around 5–6 days.
At the same time, the size of the project hasn’t changed at all, yet I’m still running into this problem now.
At the moment, I only have one suspicion — the limit starts getting exhausted extremely quickly when there is a fallback from WebSocket to HTTPS.
Because I’ve repeatedly noticed that right after this message appears, I also start seeing warnings that my limit is about to be reached.
Hello, yes it’s happening to me… just today you renewed my license and within a window of 6 hours you’ve already consumed half of the weekly and 100% daily consumption. Even if the 5.4 is a more robust model, I find this new shape disproportionate. I’ve been using codex since its inception and I had 1 times the consumption 100% and it was last week and today it has already consumed half of the plan. I use good practices with long-term projects with context management just to save money. Small tasks are using all this. I’m disappointed.
Hi pro user here,
I’ve noticed that my weekly limits seem to drop faster since the last update of codex cli.
Today I noticed that in a new chat, the limits are consumed at a normal rate until the context window exceeds 250,000+ tokens. Yes, I intentionally set it to 1 million in the Codex configuration file, because OpenAI explicitly stated in the GPT-5.4 model presentation that the model supports a context window of up to 1 million tokens.
At the same time, during the first few days after the model was released, everything worked normally, but after the later CLI updates this issue started to appear. Once the context goes beyond about 258,000 tokens, the limits start getting consumed extremely quickly. I’ll try going back to the standard context window settings and see if that fixes the problem.
I’m attaching a link and a screenshot from the OpenAI page.
https://openai.com/index/introducing-gpt-5-4/
Can anyone say for sure whether using a 1 million token context window causes the limits to be consumed much faster because it increases usage?
Also, this seems to happen specifically when the agent starts exploring a large number of files.
I had a similar issue. While using agents, it consumed 20,000 credits plus my Pro weekly limit in a single day, and it even seemed to consume credits while the agents were idle. The same amount of credits would normally last me over a month.
Did this happen to you when using agents? When did it occur? Mine happened on March 16 and 17.
Same here… weird behaviour started since the last 1-2 updates (last week or so). Hitting weekly limits within several messages - very unusual… forcing me also to switch to Claude
SAME!
Last week I coded 2x more and my weekly renewed before I get to 0.
This week it took 2 days even tho I used it a lot less!
Same! After two days using Codex CLI with Plus subscription (daily limit reached ~80%) weekly limit nearly zero.
Same here. I was cruising with xhigh for the longest time never even getting close to half the limits, then all of a sudden the usage started draining. Switching off /fast and going to “high” didn’t stop the bleeding. I don’t use any subagents.
I’ve been using Pro for four months, but it feels like on Plus now, if not worse. Needless to say, I’m getting off the next stop…
This a reminder not to get attached to one provider. Whether it’s a freak accident or not, it’s always good to have plan B and C ready.
Same here!
I noticed the same these days, the weekly limit started to drop very fast. I strictly limit the context window to 272K tokens, I use GPT 5.4 (Medium), and disable the extra-fast feature.
A simple task that used just 88K tokens, it consumes 3-4% of the weekly cap. That means that in a day I’d consume mostly 50% of the capacity, and before I had enough for a week.
I have the Plus subscription.
The week restarted yesterday with no major drainage for me now (though I’m making sure to use xhigh sparingly), but I’m already alternating with 2 different providers. I think I’ll jump on the $100 Pro (instead of the $200) if it ever comes out.
5hr — issue still not resolved
Hi all,
I’m posting because I seem to be running into the same kind of Codex usage-limit issue that others here have described.
In my case, the problem is not simply that there is a 5-hour window or a weekly cap. The problem is that the quota appears to be draining faster than my real usage would suggest, and the behaviour does not seem to line up with what I would reasonably expect from a 5-hour allocation.
I’m seeing this in a Business / Codex CLI setup, but it looks very similar to these reports:
What seems to be happening:
the weekly quota appears to run down even when I have not actually used a full 5-hour window;
relatively short or modest sessions seem to consume far more quota than expected;
failed, stuck, or incomplete runs may still be counting against available usage;
the end result is that the product becomes difficult to rely on for normal work.
I’ve already contacted support and shared evidence, but the replies so far have mainly explained the existence of fixed usage windows / weekly limits and seat billing, rather than investigating whether the metering itself is behaving correctly.
What I’m trying to understand is:
Has anyone else seen their weekly quota drain before they truly hit the 5-hour limit?
Are failed or hanging runs supposed to count fully against usage?
Has OpenAI confirmed whether this is a bug, a metering change, or intended behaviour?
If usage was consumed incorrectly, is there any process for review or restoration?
I’m not asking for unlimited use. I’m asking for transparent and accurate usage accounting, because right now the behaviour feels inconsistent with actual usage and makes Codex hard to plan around.
There seems to be a problem in 5h rate limits on VSCode extension, as they deplete after I consume 12% of my weekly usage, whereas they were originally deplete after I consume about 30% of my weekly usage.
Am I doing something wrong or did you guys lower the 5h limit?
