New limitation for ChatGPT (free tier)?

Discussion
1

Hi all, :wave:

I once in a while (perhaps a bit more often :wink:) make use of ChatGPT (free tier).
Yesterday I noticed its behaviour has changed, I´m sorry to say. :slightly_frowning_face:

Note: ChatGPT uses the GPT-4o model as per default now (at least for the free tier).

Until now it was like this:

Once you started a new topic/conversation you were able to chat along with only so many answers/responses from it until you reached a pre-defined limit.

Then a message popped up saying:

“you´ve reached our limit of messages for GPT-4o. From now on you are using GPT-3.5”

or something to that effect.
It also mentioned that at a certain time (4 or 5 hours later) it will be set back to GPT-4o.
Most importantly: you could continue your conversation.

Well, this isn´t the case anymore. :frowning_face:

Yesterday I got a different message:

“You’ve reached our limits of messages. Please try again later”.

That was it. Nothing else.
It wouldn´t let me reply any longer. :angry:

The worst thing is:

  • it didn´t revert the model to GPT-3.5 any more
  • I found no way (neither in settings nor elsewhere) to manually change the model

On https://help.openai.com/en/articles/9275245-using-chatgpt-s-free-tier-faq it says:

How do rate limits work on the Free tier?

Free tier users can use GPT-4o only a limited number of times within a five hour window. We’ll notify you once you’ve reached the limit and invite you to continue your conversation using GPT-4o mini or to upgrade to ChatGPT Plus.

Nothing like that could be found. :rage:
Just the choice of starting a new conversation.

Well, my question is:

Does anybody know of a way/workaround of manually switching the model used to either GPT-3.5 or GPT-4o mini in order to avoid such a limitation :question:

Thanks a lot in advance.

Many greetings from Rosika :slightly_smiling_face:

4 Likes
2

Hi @Rosika there are options available if you sign in.
I created a free account with a disposable email address and false details.

Not sure it works to continue a conversation though.
Anyway, if it helps, this is the choice I see when signed in.

chatgpt
chatgpt266×449 15.2 KB

3 Likes
3

Hi Gary, :waving_hand:

thanks a lot for your reply.

That´s a clever move.

In my initial post I forgot to mention that I use ChatGPT within my browser on the PC.
Your screenshot looks like it is taken from your smartphone?
In this case you would be using a dedicated app.
Am I right in my assumption :red_question_mark:

I´m asking this because it wouldn´t be the first application which works somewhat differently on the PC as apposed to the smartphone.

The thing is:

I don´t see that. I have no option of choosing the model.
Thanks a lot anyway. :heart:

Update:

In the meantime I also consulted perplexity about the problem: It said:

Switching to GPT-3.5 from GPT-4o on ChatGPT can be a bit tricky, especially if you’re using the free tier.

Note that switching to GPT-3.5 doesn’t change the default model for future responses in the same conversation. If you’ve reached your GPT-4o limit, ChatGPT will automatically switch to GPT-3.5 for subsequent responses until your limit resets.

The last statement used to be true but it isn´t any longer. That´s what caused me to post this thread in the first place.

There is no direct setting to choose GPT-3.5 as your default model for all conversations.

That´s correct in my case.

Once you’ve received a response from GPT-4o, you can switch to GPT-3.5 for that specific response by clicking on the lightning icon at the bottom of the response. This icon allows you to regenerate the response using GPT-3.5

This is for me to try.
I´m not sure if I understand it correctly.
Seems a specific response would be regenerated using another model.
But does it imply that the subsequent answers would also use this model?
And if so, would this eliminate the problem of limitation of responses?

I´ll have to try it out in my next chat. :thinking:

Many greetings from Rosika :slightly_smiling_face:

Update 2:

I just started a new chat to verify if changing the model for one answer provided by ChatGPT stets the alternative model permanently, i.e. for the following answers as well.

Turns out it does not. :angry:

  • One can reload any given answer to get a modified reply using a different model
  • The next answer however will default to GPT-4o again
  • So that´s no solution for avoiding the limitation of possible answers given by ChatGPT before being blocked for a few hours
4 Likes
4

Same here with either Firefox or Brave.
I don’t use a smartphone for it.

I asked it for “its foss information” before and after signing in.

Here’s what it returned.

Not signed in.

ifso
ifso365×559 14.9 KB

Signed in.

IF
IF376×652 33.4 KB

As you can see, there is a better result and switch model option when signed in.
No idea why it is different for you.

3 Likes
5

@Gary and all:

Hi Gary, :waving_hand:

thank you so much for your info. :heart:
That´s interesting.
So you´re using the browser version, too. There should be no difference then.

All that led me to ask ChatGPT itself about what might be going on. :wink: :

ChatGPT:

I totally get how frustrating that must be, especially if you’re in the middle of a conversation. Thanks for the detailed explanation—it helps a lot in understanding the issue.

You’re right that ChatGPT free tier users previously got downgraded to GPT-3.5 after hitting the GPT-4o limit, but now it seems that once you reach the limit, you’re locked out for a while instead of switching models.
Unfortunately, there’s no manual way to switch between GPT-4o and GPT-3.5 in the free tier. The model selection is automatic, and OpenAI hasn’t provided an option to force a downgrade to GPT-3.5 to avoid hitting the limit.

Possible Workarounds

  1. Start a New Chat:
  • The message limits apply per session, so you might be able to start a new chat instead of continuing in the same thread. This won’t restore the conversation history, but it might let you keep using the service sooner.
  1. Use a Different Browser or Device:
  • Some users have reported that switching to a different browser (or even using incognito mode) allows them to bypass the lockout temporarily.
  1. Clear Cookies & Cache:
  • Try clearing your browser’s cache and cookies, then reload ChatGPT. It might let you send more messages before hitting the limit again.
  1. Check If GPT-3.5 Works in Incognito Mode:
  • Open an incognito/private window and try accessing ChatGPT again after you hit the limit. See if it defaults to GPT-3.5 instead of locking you out.
  1. Use Microsoft Copilot (which also runs GPT-4 models):
  • If you’re locked out and need quick access, Microsoft Copilot (formerly Bing Chat) also runs GPT-4 for free and could be a temporary alternative.

Unfortunately, OpenAI has been changing how limits work, so the old behavior (downgrading to GPT-3.5) might not be coming back. If this is a dealbreaker for you, upgrading to ChatGPT Plus ($20/month) would remove those restrictions.
Let me know if any of these work for you! :crossed_fingers:

Oh well… :neutral_face:

Update:

But there seems to be some new development (possibly :thinking: ):

Note:
I had 2 discussions with ChatGPT today. In the 2nd one I asked it about the phenomon we´ve been discussiong here.
The 1st one was on a totally different subject.

Me:

You and I were discussing another topic today, in a separate thread. It´s called […].
I haven´t reached the limit for today in this thread, so I got no message whatsoever.

At the moment we are currently discussing the problem “Help with ChatGPT Issues” in this very thread, which is my second one today, and I just got the following message:

"You’ve hit the Free plan limit for GPT-4o.
Responses will use another model until your limit resets after 9:06 PM."

Do you see the difference? Today it doesn´t say “You’ve reached our limits of messages. Please try again later” but rather “Responses will use another model”.

And indeed your previous response was created by the model “4o mini”.
This time I didn´t get locked out any more. Perhaps the openai people, which are running ChatGPT, have changed their poilicy in the meantime?

ChatGPT:

Ah, I see! That’s super interesting!
It does sound like there’s been a change in OpenAI’s policy or how they’re handling the limits for free-tier users.
Based on what you’re describing, they seem to have moved away from just locking you out entirely and instead are now giving a more gradual transition.
Instead of completely blocking responses, they now revert to a lower-tier model, like GPT-4o mini, and allow you to continue your conversation with a different model.

This could be a recent tweak in how they manage the free-tier limits, which might be a response to user feedback or an update to balance server load more effectively. If that’s the case, it’s definitely a good improvement—still limiting usage but at least not locking you out entirely!

So, to summarize:

  • Old Policy: Lockout with a message like, “Please try again later.”
  • New Policy: Reverting to a different model (GPT-4o mini) and allowing you to continue chatting until the limit resets.

It seems like you’re able to carry on with a different model until the reset time. This is definitely a more flexible approach!

You’re right—this change could be very recent, and unfortunately, I don’t have real-time access to OpenAI’s updates. But it’s great to see the system is allowing more continuity. Hopefully, this is a positive shift for free-tier users!

Thanks for pointing that out! If this is the new norm, it’s definitely a big improvement. :blush:

Well, it remains to be seen if the new and improved policy persits.

Many greetings from Rosika :slightly_smiling_face:

2 Likes
6

Hmm, when they change things, just finding out what you can or can’t do is a chore.
I very rarely use ChatGPT so know nothing about their time limits/changes etc.

I guess the aim of the changes, as usual is to get the heavy users into paying for it.

From what I’ve read about AI, you get better, more up to date info if you pay.
In that case they obviously limit the info you can have for free.

Just using search engines is ok for my needs.
Prefer multiple results to peruse rather than one.

No disrespect to you whatsoever, the above is just my take on how things are.

2 Likes
7

I find all these changes off-putting.
Is there a service that we can just use without having to negotiate all these rules?

2 Likes
8

Wading through the mud of such puts me off.

2 Likes
9

If you are using AI for technical purposes, I would suggest you to check out Claude. It’s much better. ( Considering both have those message limitations now, it’s better to use claude. )

4 Likes
10

Good call @TypeHrishi

Use of AI should be selective.
It is a case of learning which one is best for purpose.
Early days to determine I guess.

3 Likes
11

Hi again, :wave:

thanks a lot for all of your replies. :heart:

@Gary :

Seems to be a plausible explanation.

No need to worry at all, Gary. Everone´s entitled to his or her own opinion.
It goes without saying. :+1:

That´s always a wise course.
In this case you might be interested in a browser extension called “ChatHub”.
See explanation below.

@TypeHrishi :

Thank you for the suggestion. I´ll be glad to try it out.

Note about ChatHub:

During my investigation concerning this topic I stumbled over a browser extension called “ChatHub”.

Here it says:

ChatHub is an all-in-one AI client that allows you to use multiple AI chatbots side by side. It supports a wide range of AI models, including ChatGPT/GPT-4o, Claude 3.5, Gemini, Llama, and more.
With a single input, ChatHub allows you to receive responses from up to six AI chatbots at once.

(bold by me)

Its main features are:

I was so intrigued by it that I simply had to try it out.

Here´s a screenshot of an easy example:

Note: easy example:
Q: What is the BBC?
I selected just 2 AI´s: ChatGPT and perplexity

kgw_ChatHub
kgw_ChatHub1585×1000 178 KB

Additional remarks:

  • the add-on is avaible for at least firefox and google-chrome (and variants of it, I guess).
  • I couldn´t get it to work with firefox (“Network error”). No idea why that was.
  • I sucessfully could run in with google-chrome.

Many greetings from Rosika :slightly_smiling_face:

2 Likes
12

I will have a try in Firefox or Waterfox.

2 Likes
13

Hi Neville, :waving_hand:

Thanks. It´s highly appreciated. :heart:

Cheers from Rosika :slightly_smiling_face:

2 Likes
14

Rosika, you are a gem.
Thank you so much for the reply.
Not gone through the links yet but will do.

As an aside I have a printer that doesn’t recognize AI.
Least of problems as of now though.

2 Likes
15

Hi Gary, :waving_hand:

thanks so much for your kindness. :heart:
I´m glad my message was of some interest to you.

Many greetings from Rosika :slightly_smiling_face:

1 Like
16

Can this be true?
If yes, that rings another bell of alarm for me…
:thinking:

3 Likes
17

I dont want to hear about them calling on AI for jury service.

3 Likes