You’ve reached the current usage cap for GPT-4, please try again after (usually 2-3 hours)
This message is ruining my day and likely causing frustration for many GPT developers. After sending 10-20 messages, I am blocked for 2-3 hours. There is no guidance on how usage limits work or how to properly test GPTs within the given limits. I have come across several topics mentioning similar issues, with hundreds of messages complaining about it. Unfortunately, I couldn’t find a single message from the OpenAI team addressing this.
Is there any plan to resolve this issue?
Do we need to find a way to work with the current limits when building GPTs?
Does anyone from the OpenAI team read this community forum?
I’ve placed a message internally at the highest place we long-term users have to get attention of OpenAI. ChatGPT policy may come from a higher place than API developer advocates.
There’s been no feedback yet on why GPT usage consumes your messages allowed per three hours at apparently twice the rate, nor why it is without mention.
Overall, AI compute capacity seems stretched thin, the basic reason why some limits are needed and signups were even paused for a month.
GPT agents (I think I’ll go back to their code name “Gizmos”…) do execute many internal model calls and use lots of data when they use uploaded files, use external actions, or use python code interpreter for data analysis.
A true “per message” usage could actually be a dozen internal AI responses for particular interactions to get a single user GPT answer.
Yes, this is frustrating, considering some test interactions may involve fewer tokens vs normal ChatGPT usage. Building GPT involves many trials to put together API interactions and fine-tuning a prompt etc. It’s a proper software development that requires plenty of debugging, so those 40 messages cap per 3 hours is laughable.
I do see how complex each interaction can be on the OpenAI’s end, but, then shouldn’t be there a “debug” mode when the GPT developer can have finer control during the development phase? I suppose it’s hard to contain the abuse of the system then.
Anyways, GPTs seem more like an “alpha” rather than “beta”, there are so many things to support proper development and I wish all the best to the team behind it!
This frustration is not only when building. Using any custom GPT (done by a 3rd party) will hit the limit real fast. Hence, I try to stick to vanilla ChatGPT GPT4, instead of custom GPTs (as useful some of them are).
So in general, having used ChatGPT steadily for a year now my experience has been the following.
There are “flurries” of this that happen every few months and can last for a week or two. In total I’d say I’ve spent about 5% of my time hitting the cap and 95% not. So it will go away.
It is likely due to the new GPT functionality and how the Knowledge and Actions are interacting with the back end. Perhaps one response that uses Knowledge counts as “2” and a response that uses Knowledge and Actions counts as “3”. That may be off-base but it feels right given my experience.
Switching to GPT-4 makes you less likely to hit the cap. Indeed there have been times where I hit the cap with a GPT and it still allowed me to start a new thread with GPT-4 (vanilla ChatGPT). For clarity, I have hit the cap with GPT Builder multiple times, opened a new window, and used vanilla GPT-4 and was not cap-restricted.
Version 3.5 is actually quite powerful and not capped at all. You can’t do everything you can with a GPT or GPT-4 but there is a lot you can do, and fast, with it. If you batch some other tasks and interact with 3.5 for a bit that can be a productive way to pass the downtime.
It can be frustrating for sure.
In general, I really like the idea from @former.token and @_j to make it easy for Custom GPT developers to request a higher cap, or make GPT creation “cap free”. It’s a manual process anyway and is truly a developer workflow that takes cycles. Hopefully OpenAI can take notice of that request.
Oh well, the only way now is to plan it to extreme detail.
Debug your API separately with Postman, webhook.site, or anything your heart belongs to.
Do all edge cases, remove authentication from GPT integration to make it easier.
Spread your 40 messages as evenly as possible over your working hours. Plan for that cap disruption so you have to work on something while it’s getting restored.
To be honest, this is an entirely new way to develop software, and the messages cap is the least problem you will encounter. My 25+ years of web app development do help quite a bit, but this is another tech leap.
Any solution found so far ? Please update or if the Developer Community is looking , should come forward and enlighten us as i was just discussing story of my life with GPTPlus4 and just when it got interesting and i took a pause and then resumed my chat , the Cap issue emerged and now i am feeling absurd and weird .
Yes, this is incredibly frustrating. Funny thing is that I only encountered this only after subscribing. I don’t even use 40 messages. After about 10-20 messages, I am also blocked for 2-3 hours - which I agree, makes it very difficult to build GPTs. After reading these responses now I probably agree that OpenAI includes all prompts when building the GPT in the 40 limit. What I don’t understand is why they would include a limit for paying subscribers. By all means, set a limit for free users but have more tolerance for paying members!
