In regard to your last post, you’re on a plus tier or a pro tier?
Because it doesn’t sound like you’re actually paying for 7 days…
It sounds like you spent 20 dollars and want your 5-dollar-per-week budget to cover 7 days of round the clock codex usage?
I am a Plus subscription user.
I just want to cry anyway, 
I understand that the problem remains a problem, I’m spending time 
Yeah, so you pay the lowest tier and think that it would cover around-the-clock usage for a week.
Why would there even be professional tiers then, for people who use it around the clock seven days a week?
It was nice chatting, but this conversation has no point anymore — I’ve said everything I wanted to say.
Keep your opinions, I’m not interested.
I wrote all this for people like me who face the same problem and waste time figuring out what to do next. If it saves someone even an hour — that’s enough. It wasn’t for you.
I honestly don’t understand what your angle is here — calling users idiots while acting like the smartest guy in the room. Not my problem. 
Good luck to you. 
Never called you names, just pointed out that you were making many long posts about how you didn’t have enough codex time when the solution would be to pay for a plan that suits your needs, friend.
To stay constructive. I noticed that today my option to see the remaining time for the 5h and the weekly disappeared.
I’m a plus user and I wonder, is it because of all the turmoil that they deactivated the limits until they fix it or something alike?
This was noted in a similar topic
Updated on: May 20, 2026
https://help.openai.com/en/articles/20001106-codex-rate-card
I’ve tried hard to adapt to the new Codex limits instead of just complaining about them.
I wrapped my use cases in skills, optimized my AGENTS.md, added agent-specific npm scripts with limited output, narrowed task scopes, and tried older models like GPT-5.4 mini with low effort/intelligence settings.
But even then, the first message of a brand-new Codex chat can instantly consume around 10% of the hourly limit. That makes the limit feel unpredictable and very hard to work around, even when using the product carefully.
I used to promote Codex to colleagues because the pricing felt reasonable compared to Claude Code. I can’t honestly say that anymore. For the first time, moving more of this workflow to local models like Qwen actually starts to make practical sense.
I’m not asking for unlimited usage. I’m asking for predictable, usable limits and clearer visibility into what a task is going to cost before it consumes a significant part of the allowance. The change in token usage changed; this suddenly made existing accounts less usable. I am hoping someone from OpenAI reads this and reconsiders.
Agreed. It’s extremely clear that the rate limits are (for us) arbitrary. A percentage of a moving target.
OpenAI currently has the most generous rate limits for cloud coding harnesses, but I think (by design?) that hiding the actual token usage is pushing people into unreasonable usage patterns.
Needless to say, I think the only predictable pattern is a continuous decline, making credit purchases more appealing. Consumers/hobbyists were a stepping stone.
This is the wey. Start bootstrapping yer own harnesses and models before yeself is found trapped inside a rotting garden dug too deep to ever crawl out of
This is a cool place to monitor how terminal harnesses are doing.
Hey all, wanted to add a bit more context on how Codex usage is counted and what’s worth checking when usage feels higher than expected.
Codex usage is not based on “number of prompts” or wall-clock time. It depends on the model, task size, repo/context size, local vs cloud execution, and token usage. The current Codex pricing/rate card is here.
A few things that can make usage move faster than expected:
- Large repos, long threads, big `AGENTS.md` files, MCP/tool context, and repeated compactions can all increase input/cached/output tokens.
- Model choice matters. For example, GPT-5.5 is more expensive than GPT-5.4 / GPT-5.4-mini on the Codex rate card.
- Fast mode uses more credits: https://developers.openai.com/codex/speed
- Cloud tasks, code review, subagents, and automations can keep doing work in the background. Automations are documented here: https://developers.openai.com/codex/app/automations
- Memories can also do background work. Codex may update memory files after a thread has been idle long enough causing this impacts your usage. You can review/control this via Codex settings or `/memories`, and memory files live under `~/.codex/memories/`.
So if your app looks “idle” but usage still moved, the first things I’d check are:
1. Codex Settings → Usage & billing.
2. Run `/status` in an active CLI session.
3. Check whether any cloud tasks, code reviews, automations, subagents, or background terminal runs are still active.
4. [Most common]Check whether Memories are enabled with `/memories` or in Codex settings.
5. Compare the model/reasoning/Fast mode setting used for the run.
If something still looks wrong, you should submit a support request to us at support@openai.com and our team can look into your usage. The most useful support report is: plan type, model, local/cloud, exact timestamp + timezone, before/after usage screenshot, and whether Memories/automations/subagents/cloud tasks were enabled. That gives the team enough info to separate “expected token/background usage” from a real accounting issue. We cannot attain those information publicly so we request you to please use the support path. We appreciate and thank everyone for coming together here and engaging the community.
I currently use two ChatGPT accounts:
i noticed something very strange in codex while using between these two accounts.
I asked the exact same question using the same model, gpt 5.5 medium,
the prompt was very simple, something like:
"change the title on the frontend to … "
on my personal plus account, this kind of general question usually consumes less than 5% of the 5-hour usage limit
However, on my Business account, the same type of request easily consumes around 10–20% of the 5-hour usage limit.
For slight more complex tasks, my plus account usually consumes around 10% ish,
but my business account consumes 30–50% or even more everytime from one single question.
this has happened multiple times, not just once.
I already contacted OpenAI support, but the replies were copy-and-paste responses, or no response at all LOL like fully ignore my message.
the support team definitely not know whats going on, felt like talking to npc and when i asked about tokens burning so fast, they ignored (Hello Lance 
yup you ).
has anyone else experienced this issue with codex usage being much higher on a business account compared to a plus account?
I would appreciate any explanation
I’ve noticed the same issues. I had been on the codex personal plan €20 and never even hit my 5 hour limit or whatever it is, I also use claude too and would hit that but be nowhere near my weekly limit. My all time token usage on that claude account which is a couple months old is 4.4m tokens, the enterprise codex account I have has apparently consumed 802m tokens this month, which I can’t even see how it’s possible. Obviously I work as a software engineer, but I’m like a 1.05x engineer, not one of these super users. I also got no reply from open ai support
Are these requests in threads that have a lot of previous requests?
I ask that because I recently had to transplant a project that I rebuilt into it’s own repo.
I found out that the size of the repo effects your credit usage as well.
One of several key things that helped me reduce credit spend was looking out for many artifacts that are being drug around with the session.
If you’re using the desktop app and it’s not full screen, just move your mouse to about the middle of the right edge, and move it up along the edge about 20%…
Check to see how many artifacts you’re carrying around.
13 is a pretty big number I’ve got 1 session that drags over 60 docs around as artifacts, it seems like this has an effect on overspending too.
Be mindful of what docs your bot needs in order to have the map it can use safely…
Ask it to aduit the docs it needs for future work..
When say, an engine is completed and you’re building around it put it inside of the agents.md that it shouldn’t be edited unless absolutely required…
I’ve discovered there’s a lot of things that can bloat your consumption, but one of the main things is not letting automation check things you can do with a visual test yourself.
I believe I may have encountered an issue with Codex usage limits.
I started a coding task and then stepped away from my laptop. During that time, my screen turned off and the computer went to sleep. When I returned later, the task had still not completed, but my available Codex usage had been exhausted and now shows a reset date several days away.
The task itself was relatively small, so I was surprised to see all available usage consumed without a completed result.
It’s possible the task became stuck or continued running in the background after my screen turned off. If that’s what happened, it may be helpful for Codex to detect situations where a task appears stalled, disconnected, or inactive for an extended period and automatically pause or stop execution rather than continuing to consume usage indefinitely.
Nah man, you ain’t alone. I have been facing the same issue over here
Same issue here – Pro 20x, dropped from ~60% to 0% today with barely any usage. Clearly not an isolated case. @Tristan323 any update on Case #09851164?
