It really needs to be re-examined because it sets limitations that were not there previously. My entire workflow that managed to work without those limitations and still follows the guidelines of fair use according to the plan is now being throttled by this decision. Please re-consider. Because if this continues much longer, I will be looking at other options very soon.
No it is not. It’s an engineering safeguard to help maintain service for everyone. Just like any rate limit for any SaaS.
If you don’t like it you can upgrade to Pro which doesn’t have as many users so the regime is a bit looser atm.
I think top-tier Codex is more of a bonus like a paid trial than a guarantee
You get access to Chat, Work and Codex which is perfectly fine with like 1 Sol Medium for planning/review and you can run Luna XHigh it eats about 50% of usage every 5 hours meaning the rest can go on Sol Medium to first set up a plan and then to review. Analyze plan give it a long running task and review at end
terra x high for planning is perfectly acceptable .. you need sol for crazy stuff that requires math that can’t simply be googled..
Passer à l’abonnement Pro constitue effectivement une solution, mais cela ne répond pas vraiment au problème soulevé.
L’offre Pro est principalement adaptée aux utilisateurs intensifs — notamment ceux qui emploient Codex dans un cadre professionnel et peuvent rentabiliser son coût. Pour un particulier qui utilise Codex pour apprendre, créer une application personnelle ou mener ponctuellement un projet, la différence de prix avec l’abonnement Plus est difficilement justifiable.
La question n’est donc pas de nier que Pro offre des quotas beaucoup plus élevés, mais de constater qu’il existe un écart tarifaire important entre Plus et Pro. Tous les utilisateurs qui atteignent occasionnellement leur quota n’ont pas pour autant besoin — ni les moyens — de souscrire une formule destinée à un usage professionnel intensif.
Une formule intermédiaire consacrée à Codex, des crédits supplémentaires plus accessibles ou un système de renouvellement moins contraignant constitueraient probablement des réponses plus proportionnées pour les particuliers.
Disagree. If you have not read what I wrote, re-read it again. Within the plan you have, there is a ‘budget’ of tokens you can use. That budget was distributed weekly over 4 weeks. That worked. The problem with the 5-hour limit is it creates this artificial throttle that can be disruptive to any ongoing workflows. Budgeting use for a week is a lot easier than budgeting for five-hour intervals.
But of course your solution is to “upgrade”. Yes, “give us way more money for the same service you had before” .
There are simply a lot more Plus users. Hence the need for more rate limits to keep control of system availability. The amount of users they are having to cope with is unprecedented …
If you don’t want to believe someone who has worked on IT systems for ~25 years you can instead listen to the man himself:
I think a 50 dollar plan would be a nice compromise
More than 20 but not as expensive as 100
Or you can simply just follow the money.
I don’t trust such metrics…
I hear ya. They should call it the 45 minute, not 5 hr lol. But i guess what we are all doing is finding the right path. Sure they are making billions, but its all new, and its all us making it work. In the early days, people paid for email accounts.
This used to work as expected, but at some point, I’m not sure exactly when, the behavior seems to have changed.
OpenAI’s current documentation says:
What happens if I reach a usage limit while Codex is working?
If you reach a usage limit during an active turn, Codex can continue working on that turn, subject to fair-use limits. After that turn, check the Codex usage page or the limit banner for the options available on your plan, such as adding credits, applying an available reset, upgrading, or waiting for the limit to reset.
Previously, if a Codex task was already running when my usage limit reached 0%, the active task would continue until it finished.
Recently, however, the behavior appears to have changed. An active task now seems to stop immediately when the usage limit reaches 0%.
I’m not sure whether this is a bug or an intentional policy change. I also don’t necessarily object to stricter usage limits themselves.
What I’m wondering is whether this change is intentional, and what exactly “subject to fair-use limits” means in this context.
If the behavior was intentionally changed so that reaching the normal usage limit now also causes the active turn to stop immediately, I think the official documentation should be updated or clarified to reflect that behavior.
As it currently reads, the documentation suggests that an already-running turn may continue after the normal usage limit is reached, subject to some separate fair-use constraint. If reaching the normal usage limit now also immediately reaches that fair-use limit, then the current wording seems somewhat misleading and could probably be clarified.
happens to me too since yesterday, so annoying
This is the truely disruptive part of the change, I don’t mind having 5h quotas we used to have them before. The disruptive part was that they removed the finish task to completion part. If they are going to do that they should probably just change “subject to fair-use limits” to “if you are lucky”. For programming workflows it’s not disruptive as you can just stash the changes and revert to the last git checkpoint, then pop them off the stack and resume when your quota returns. But for jobs involving interacting with live systems over command line, it’s incredibly disruptive as you have no guarantee that the system will be left in a controllable state, unless you always stop when you have 5% 5H quota left using a steer command telling it the state you want it to pause at, this requires you to be present watching the quota as it works. I wouldn’t feel so bad about it the pay as you go credits weren’t priced so differently as I could just buy a handful of those to take jobs to completion (which is what I assume they want you to do) but that has the potential to get expensive fast…
I’m a paid user trying to use ChatGPT Work for real website and project work. The “5-hour limit” is not five hours of usable working time—it is an unpredictable compute meter that can run out during a task even with plenty of weekly usage remaining.
Tonight I stayed up an extra three hours waiting for the reset so I could finish a website fix. It completed the fix, but I was blocked again before I could send the one final command to finish the job.
Higher plans giving “5× more usage” do not solve the actual problem if the same rolling five-hour shutdown still interrupts work. Even purchasing the $100 Pro plan does not fix this issue—it gives more capacity inside the same five-hour throttle, rather than reliable continuous access.
I have also spent an additional $80 in credits trying to keep working. Paid users should not have to keep buying credits just to avoid being shut down in the middle of important work.
Please offer predictable daily/monthly capacity, a real time/usage forecast, or pay-as-you-go continuation instead of forcing paid users to stop mid-task.
At minimum, give paid Work users a realistic 8-hour workday allowance. Five hours is not enough for people using this as a serious work tool, especially when the meter can drain far before five actual hours of work. For people who work long days or 50–60-hour weeks, the current system does not feel like a professional plan. This needs to be fixed.
It is really predictable. For 20$ you get a tiny bit of it. That is as predictable as it gets.
The 200$ limit doesn’t even last a full day when you go full mental and use astra on ultra with tons of subagents.
You are going full extreme to try to illustrate a point–it doesn’t work. What people notice is that their regular use that used to be acceptable and not hit any real limits prior to the recent changes is now becoming real limitations. And OpenAI is trying to gaslight us by stating that they have not done any kind of nerfing to the usage tokens. Let me guess why…hmmm…maybe because they want you to pay more money for the same allowed usage you had before?! Oh, no, of course they can’t be doing that… Spoiler alert: they are.
What do you mean by that? They give you access to a private jet for 20$ and you complain that you can’t use it 30 days per month?
The plus plan is for testing not for working.
Terrible analogy. This is not a “private jet”. It is a service they provide. And the amount to pay for that service was agreed to by both parties (the user and service provider). Now they decided “we are going to reduce what you used to have before and if you want that same service, you need to pay way more”. And, no, the Plus plan is not “for testing”. That is what the free plan is for. The Plus plan IS for working. The Pro plan is for large teams and companies.