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
Interesting product which seems to have become a focus for what I see as the central theme of AI tools right now which is the sustainability of the business model in the face of ever-more resource hungry tools and potentially insatiable demand!
My own problem is simple and unimportant but for what it’s worth I started to trail Codex following a prompt in ChatGPT and was progressing well with a small project to evaluate its capabilities but then hit a resource cap which means I have to wait for a month under the free access model before I get to find out enough about what it can do to justify a purchase. Looking at these pages, it seems that even if I were to pay for capacity, I might continue to have the same problems, albeit on ‘the next level up’ so to speak. As it is, I’m happy to wait for a month before continuing with my trail of Codex’s capabilities but I wondered what people thought about the wider issues here? Is AI an unsustainable bubble or if not, how will it shape up?
This is ChatGPT’s summary of the situation I encountered:
“I trialled Codex on a genuine project involving cataloguing a large photograph collection. The experience was promising, but I reached the free usage limit before I had progressed far enough to determine whether Codex was actually suitable for the task. The current free allowance seems sufficient to demonstrate the interface, but perhaps not sufficient for evaluating a non-trivial workflow.”
Previously, the five-hour usage limit for Codex on the Plus plan was at least enough to go through a few rounds of prompts.
Now, after just one prompt, the five-hour limit is gone. It feels like the processing time has become longer, but the actual amount of code written has not increased. In other words, it has become slower while consuming the usage limit faster.
Before, Codex could directly modify files, and it would clearly show which files were changed and how many lines were modified.
Now, it mostly changes content through commands, making it hard to see what was actually changed.
The sandbox is already set to sandbox_mode = "workspace-write", and the project is also set as trusted.
The frequency of prompts requiring manual authorization has also increased significantly.
Hi everyone,
I’m currently on the ChatGPT Pro 5x plan and I’m experiencing severe Codex usage drain on both my 5-hour quota and weekly quota.
What makes this confusing is that my usage was normal for the first few days after I bought Pro. Then, after OpenAI said the Codex usage-limit depletion issue was fixed, my quota started draining much faster than before. This happened after the reported fix, not before.
I was planning to upgrade to the 20x Codex usage tier next month, but now I’m hesitant because I don’t know whether this is a temporary issue or not.
I’m not running unusually large tasks compared to before. The same type of Codex work that felt normal earlier now appears to consume a disproportionate amount of both 5h window and weekly usage.
Has anyone else on Pro experienced this after the incident was marked resolved?
I would especially like to know:
-
Whether others still see abnormal quota drain after the reported fix
-
Whether this could be account-specific metering or rate limiting
-
Whether Pro users were moved to a different Codex usage calculation recently
-
Whether OpenAI Support has confirmed anything for anyone affected
I wanted to ask here because the issue directly affects whether upgrading to 20x is reliable.
I’ve swapped back to 5.4 medium when using Codex, and it’s been way better for my workflow. With 5.4, I rarely hit the rate limits, but the second I use 5.5, I lock myself out in two hours and max out my weekly cap in three days.
In my opinion, 5.5 medium is only marginally better for coding tasks. By just using plan mode on 5.4 and being a little more specific with my prompts, the gap disappears. For me, not getting locked out is worth way more than a 5% difference in thinking quality.
My 5 hour limit is still depleting very quickly. It took 10 minutes to deplete while doing a fairly heavy task, but its still very unusual to deplete that fast.
I’ve been on Plus ever since, I usually hit the 5 hour limit within 2-3 hours.
OpenAI already “resolved” the issue a couple days ago, but still no improvement.
Edit:
I switched to 5.4 model and usage is somewhat normal
Title: Codex usage limits feel inconsistent after 5-hour reset — could ChatGPT Web usage affect Codex limits?
Hi everyone,
I wanted to ask if anyone else has noticed similar behavior with Codex usage limits recently.
First of all, I really like Codex and use it a lot. I have been using it for quite a while and honestly do most of my coding/workflow through it. Depending on my current workload, I switch between Plus and the higher-usage plan, so I am not against paying for higher usage when I need it.
However, recently the usage limits have felt very inconsistent and hard to understand.
What I noticed:
-
After exhausting my Codex session limit and waiting for the 5-hour reset, I sometimes do not seem to get a full session back.
-
In some cases, the next session already appears partially used, for example only around 40–50% remaining.
-
Sometimes I hit back-to-back 5-hour waits shortly after a reset.
-
This seems to happen even during periods where I am no longer using Codex directly and only use ChatGPT Web.
-
I am wondering whether ChatGPT Web code generation or Python execution might somehow count toward Codex usage, or whether the usage tracker/reset is delayed or bugged.
I tried to test this with a simple Python QR code generator in ChatGPT Web. While my Codex usage was limited, the Codex usage did not visibly change immediately. But after the 5-hour reset was over, part of my new session limit and some weekly usage already seemed to be consumed, even though I had only used ChatGPT Web during that time.
I am not saying this is definitely how it works, but something feels different from before.
My questions are:
-
Does ChatGPT Web coding or Python execution count toward Codex session or weekly limits?
-
Can a new 5-hour Codex session start already partially consumed because of delayed usage tracking?
-
Were there any recent changes to Codex limit calculation for Plus or higher-usage plans?
-
Has anyone else experienced back-to-back 5-hour waits or partially depleted sessions right after reset?
I understand that limits are necessary, especially for agentic coding tasks. But right now it feels difficult to predict what actually counts toward Codex limits, and the reset behavior is not very transparent from the user side.
Would appreciate any clarification or similar experiences from others.
I have noticed something similar recently.
In my case, I have seen GPT-4-mini usage get exhausted much faster than before. I reached the 5-hour limit in roughly one hour with a workflow that, until around two weeks ago, had never brought me close to the limit.
My main use case is code-review work: I use GPT-5 in ChatGPT for reviewing code and related development tasks, rather than only using Codex directly. Because of that, I also suspect that the token-accounting policy may have changed, or that some types of usage across ChatGPT and Codex may now be weighted or aggregated differently than before.
I cannot confirm whether ChatGPT Web usage technically counts toward Codex limits, but the practical behavior has definitely felt different recently: limits are reached earlier, resets can appear partially depleted, and it is harder to predict how much usage is actually available.
More transparency around which models, tools, and workflows contribute to the 5-hour and weekly limits would be very helpful.
Ok, i think also the github code review was wasting toons of tokens. i will see if without it everything works fine again.
Follow-up after checking my Codex session logs
I checked the exported JSONL logs for the 12 Codex Desktop sessions I ran on July 4–5.
All 12 sessions explicitly report:
- Model:
gpt-5.4-mini - Reasoning effort:
high - No visible
gpt-5.5calls - No visible subagents, spawned agents, or nested tasks
However, the account usage dashboard for July 5 shows:
gpt-5.4-mini: 19 turnsgpt-5.5: 31 turnsgpt-5.4: 2 turns
I am attaching the screenshot because this creates a real attribution problem.
The Codex Desktop logs themselves do not explain the GPT-5.5 usage. I do use GPT-5.5 in ChatGPT for manual code review, planning, and product work, so one possibility is that the dashboard aggregates ChatGPT Web, Codex Desktop, Code Review, or other surfaces into the same usage picture. Another possibility is that some workflow uses a different model behind the scenes without exposing that clearly in the local Codex session logs.
I cannot prove from this that GPT-5.5 ChatGPT usage directly consumes the same Codex quota. But I can confirm that, on the same day, my local Codex task logs show only GPT-5.4-mini while the usage dashboard records substantially more GPT-5.5 turns.
That makes it very difficult to understand why a 5-hour limit is exhausted, especially when the visible Codex work appears to be using a smaller model.
It would be very helpful to have:
- a breakdown by surface: ChatGPT Web, Codex Desktop, Code Review, Cloud Tasks, etc.;
- confirmation of which surfaces share the 5-hour and weekly pools;
- any model or task weighting applied to limits;
- visibility into whether subagents or background model calls are counted;
- an explanation of what activity contributed to a specific limit reset or exhaustion.
Without that, users cannot reliably tell whether they are consuming Codex capacity through coding tasks, ChatGPT Web reviews, automated code review, or some combination of all of them.
$200 Pro user issue: Over the past two weeks, my weekly quota has been consumed at an extremely high rate—almost 40% or more per day, depleting it completely within just two days. This was never possible before. Additionally, I usually have over 50% of my five-hour quota remaining each time, yet my weekly quota is being used up so quickly. I’m unsure why this is happening and would appreciate a response.