I wanted to share a Codex usage issue I’ve been experiencing since 10 May 2026, in case other developers are seeing the same thing.
My Codex limits appear to be getting consumed unusually fast. In my case, after only around 1 hour of light Codex usage, mostly using GPT-5.3 Codex / mini-level models, I was already close to hitting usage limits.
This feels abnormal because earlier I could use heavier models like GPT-5.4 with extra thinking for much longer, and the usage still stayed under control. Now, even lighter tasks seem to consume the limit extremely quickly.
The concerning part is that the usage-tracking UI also appears buggy. The section/function that should clearly show usage details is not showing properly or seems hidden, which makes it harder to understand what is actually consuming the limit.
I’m not saying this is definitely intentional. It could be a genuine bug, a recent change in how usage is calculated, or some issue with Codex sessions/workspaces consuming usage unexpectedly. But from a regular user’s perspective, it is not reassuring when:
usage is consumed much faster than before,
the usage tracking UI is not reliable,
and users are being pushed toward buying extra credits without clear visibility into what caused the limit consumption.
I’ve also seen people online and heard from friends/colleagues that they are noticing similar behaviour recently. That makes me think this may not be an isolated issue.
Has anyone else experienced this since around 10 May 2026?
Specifically:
Codex weekly/daily limits getting consumed too quickly.
GPT-5.3 Codex or mini-level usage counting much heavier than expected.
Usage dashboard or tracking UI missing/buggy.
Limits being hit even without heavy or long-running tasks.
I’ve already contacted OpenAI Support and asked them to investigate the usage calculation.
Would appreciate hearing if other developers are seeing the same issue, especially regular Codex users who actively track their limits.
I’m using Business Plan and also finding abnormal token usage. Initially I thought this was due to an issue on my part, but even after explicitly trying to minimize token usage it’s still draining quickly. I’ve hit 5h limits twice today, and now only have 40% of my weekly left.
Honestly pretty disappointed with the OpenAI support experience right now.
I submitted:
screenshots,
timestamps,
usage inconsistencies,
UI bugs,
and even video proof showing the Codex usage tracker disappearing entirely.
I also explained the logical inconsistency where my weekly limit suddenly jumped to ~90–94% while the 5-hour limit was still much lower, which clearly suggests something abnormal with tracking/accounting.
Instead of actually investigating the issue, checking backend logs, or escalating it to engineering, I’m mostly getting generic copy-paste responses explaining how token usage works — things that are already available in the docs and chatbot responses.
That’s the frustrating part. I didn’t contact support to be told basic documentation again. I contacted them because there appears to be a real bug or accounting issue.
What makes it worse is:
other users have reported similar Codex tracking issues in the past,
OpenAI has previously acknowledged tracking problems publicly
As a paying user who actively tracks usage, this honestly makes me disappointed. Support should investigate customer-specific evidence instead of responding like another automated chatbot.
Has anyone else here managed to get an actual engineering escalation for Codex usage/accounting issues recently?
i don’t know if this is relevant for your situation but what i found solved my token drain was removing the suggestions setting or disabling the suggestions in codex app.there is another setting auto review prs or something on the codex settings on the website disabled both of them
I’m also seeing the same issue of unexplainable fast drains of both my weekly and 5 hour quotas.
Since ~May 10, I’m seeing that I can only get 2 basic prompts executed before my 5 hour quota is completely drained. There is also ~15-20% drop in my weekly quota after the same 2 basic prompts.
I love Codex and have been a loyal user, but progress on my projects has stalled as a result of this recent change and I’m considering other options to maintain progress.
Codex broke the quota system back in October 2025, rendering Codex unusable for basic development for several weeks. It was terrible to see them do that then, but they heard the message and restored reasonable quotas.
I hope we can just wait a few weeks for the Product Manager(s) at Codex to see that the recent quota change that occurred ~May 10 is not serving their customers and likely hurting their internal metrics.
I understand you. Today I was genuinely very angry and wrote about my frustration with Codex. The main issue for me is that the limit calculation system feels unfair and misleading. To use a simple analogy: the service suggests that I can work with code for up to 5 hours, but in practice I got around an hour, or even less, of real useful work. Technically, “up to 5 hours” can also mean 5 minutes, but from a user’s perspective this feels like dishonest wording. What disappoints me the most is how quickly the limits run out. Because of that, Codex has been almost the most useless program on my computer for an entire week, even though I am paying for it.
Is it really not possible to reduce the speed, limit the model’s power, or at least keep access to a very basic model so that paying users can still use coding features? For example, I would like to use a weaker model for tests, project audits, or simple checks, so I do not hit the limit so quickly and can continue using the program. Overall, I am still angry and disappointed. Next month, I will most likely try other tools, because I do not want to pay for a service that effectively does not work when I need it.
I think there’s an important misunderstanding here regarding the limits.
The “5h limit” is not a limit where you can use Codex for only 5 hours per day. It’s actually a token/computation quota measured over a rolling 5-hour window.
I initially misunderstood it the same way, especially because it’s displayed next to the weekly quota, which makes it look like a time allowance rather than a usage window.
I’m not entirely sure why OpenAI chose this system, but my guess is that Codex enables extremely heavy workloads compared to normal chat usage. With the recent promotional/free-access period, a huge number of users probably started running very intensive coding sessions simultaneously.
That likely created enormous compute demand on the servers, especially since Codex tasks can consume far more resources than standard conversations.
I do agree that labels like “daily limit” and “weekly limit” would probably be clearer and less confusing for most users, though.
Same problem, but for me, the issue was since the weekly usage recharge on May 16th I think.
Before that, every 20-min task under gpt-5.5 cost about 5% of the 5-hour limit; now it costs about ~25%. Bascially I can see 1% decrease every minute or even every 30 seconds when the context window is full. I used up the 5-hour credit in about 1 hour.
Hello? Will there ever be someone from OpenAI looking into this? I don’t have a problem with slightly fluctuating quota usage, this has to be expected. But for a few days now the token usage absolutely exploded:
Concrete example: with gpt-5.5, a single, very simple message + reply, with no (significant) tool calls at all, consumed about 14% of the 5-hour limit on a Business Plan account. After switching to gpt-5.4, 3-4 messages of similar simplicity, again without tool calls, consumed only about 1% total (as expected and as has been the norm in recent weeks, even with gpt-5.5).
That is unsustainable. We need an answer or clarification ASAP.
Someone on the forum told me to just pay more They say the problem is that I bought a plan that’s too cheap. Apparently, my ambitions are too big for $20
But then I have to ask: what are you actually supposed to build on this plan if even users on the most expensive tiers are complaining about limits?
I tried different tactics: shortening my prompts, removing extra spaces and line breaks, and asking for short answers without explanations. Yes, it saves some usage, but not enough to actually solve the problem.
Now I’m looking at Antigravity 2.0: Google has introduced its own agentic tool, similar to Codex/Claude Code. Based on user feedback, it also has quota issues on the more expensive models.
One prompt asking Codex desktop app if it could see a tab open on my browser cost 6% usage now (worked for 53s) wtf. I rarely run into usage issues before, now my 5h limit usage only lasts around 1 hours (one time 10mins all gone). I thought the plus tiers is not enough so I bought additional business tier for two seats, felt like magic.. pooof and everything’s gone.. no more usage for you sucker.
One practical way to debug this is to separate three counters that often get collapsed together: prompt text size, tool/runtime activity, and verified progress. A short prompt can still become expensive if it opens browser/tool loops, retries after failures, or keeps re-reading large context without producing a durable change.
The usage UI would be much easier to trust if each Codex run exposed a small receipt: model used, wall time, tool calls, context/read volume, whether files/tests changed, and the stop reason. Then users could tell the difference between a real metering bug, a workspace-specific loop, and a normal higher-cost task.
They say the problem is that I bought a plan that’s too cheap. Apparently, my ambitions are too big for $20 
