I’m a Google AI Pro subscriber who primarily uses AI Studio.
I’ve seen other users reporting similar behavior, but I wanted to document my own case separately because I have some additional comparison data.
Over the last several days, my practical AI Studio usage has changed dramatically:
- Previous Pro experience: generally around 40–50 generations per model before hitting the limit, sometimes more on Flash models. I specifically remember reaching roughly 50 generations on 3.1 Pro.
- Current Pro experience: often only around 15–20 generations in my normal long-form workload before I hit the daily limit.
- Current non-subscribed account test: roughly 10–30 generations depending on the model, using deliberately tiny prompts.
Several older models that I previously used as fallback options are also now marked for paid usage.
These are my own observations, not official Google quota figures.
But Google explicitly advertises increased usage limits in Google AI Studio as a Google AI Pro benefit.
So my question is simple:
Is this a quota / entitlement / provisioning problem, or is this actually the new intended Google AI Pro experience?
I am not asking for unlimited compute.
I am asking why a paid benefit that previously felt meaningfully usable now appears remarkably close to the non-subscribed experience — and why paying users have to manually count prompts just to figure out what their subscription actually provides.
What I’m seeing now
I mainly use AI Studio for long-form creative writing and roleplay.
Recently I’ve used 3.1 Pro, 3.6 Flash, 3.7 Flash, and 3.8 Flash. In my actual workload, I now repeatedly end up around roughly 15–20 generations before hitting the daily wall.
Not 100.
Not 50.
Around 15–20.
And then, effectively:
“Thanks for using AI Studio. See you tomorrow.”
At first I wondered whether my long contexts were simply consuming some hidden compute or token allowance much faster than normal.
So I tested that possibility.
I tested a completely non-subscribed account
I used a separate Google account with no Google AI subscription at all.
For this comparison, I deliberately used extremely short prompts so that long context windows, huge token counts, or complex reasoning would interfere as little as possible.
What I personally observed was approximately:
- Gemini 3.1 Pro: ~10 generations
- Gemini 3.6 Flash: ~20
- Gemini 3.7 Flash: ~15
- Gemini 3.8 Flash: ~30
Again, these are not official quota numbers.
I simply counted generations until AI Studio stopped allowing further use.
But that is exactly why the current situation seems so strange.
I’m paying roughly $20/month for Google AI Pro, yet a completely unsubscribed account appears to receive usage in roughly the same ballpark.
Maybe 10 generations becoming 20 technically qualifies as a 100% increase.
Mathematically, congratulations. It doubled.
In practical use, however, 10 and 20 generations have another important property in common:
both disappear very quickly.
There is a major difference between an increase that looks impressive as a percentage and an increase that actually provides meaningfully more usability.
Google explicitly presents increased usage limits in Google AI Studio as a Google AI Pro benefit.
If the practical difference from a non-subscribed account is only a handful of additional generations, then technically satisfying the word “increased” is not the same thing as providing a meaningful paid benefit.
I actually remember hitting the previous limits
This is not based only on a vague impression that “AI Studio used to be more generous.”
I have actually used AI Studio all the way to its quota limit on many days before.
From my own experience, individual models generally lasted somewhere around 40–50 generations, while Flash models could sometimes go beyond that.
Even 3.1 Pro — not Flash — lasted roughly 50 generations in my experience before I finally hit the limit.
Again, I am not claiming that 50 was Google’s officially published quota.
I am describing what I personally experienced after repeatedly using the models until AI Studio stopped me.
That distinction matters.
But it also makes the current experience much harder to dismiss as simple perception.
Going from roughly:
40–50 generations per model
to:
15–20 generations
is not a subtle adjustment.
That is not “a little tighter.”
That is the usable allowance dropping to roughly one-half — and in some cases close to one-third — of what I was accustomed to.
And for 3.1 Pro specifically, going from roughly:
50 → 15–20
is…
impressive.
Just not in the direction I expected from a paid plan advertising increased usage limits.
So yes, I remember the previous limits because I actually hit them.
That is exactly why the current change feels so dramatic.
And now the fallback models are becoming paid too
There is another recent change.
Several older models that I previously relied on as fallback options are now marked for paid usage as well. lol
Previously, if I exhausted the included allowance on one model, I could at least switch to an older model and continue working.
It wasn’t ideal, but it kept AI Studio usable.
Now that fallback route is becoming much more limited too.
So, based purely on my own experience:
Previous Pro: roughly 40–50 generations per model, sometimes more on Flash
Current Pro: often around 15–20 generations in my normal long-form workload
Current non-subscribed account: roughly 10–30 generations in my short-prompt test
Older fallback options: increasingly requiring paid usage
Apparently this is what “Pro” looks like now.
To be clear, Google can decide which models are included, which require pay-per-request usage, and what the quotas should be.
I am not arguing that every model should be free or unlimited.
What I am questioning is the combination of all these changes while “increased AI Studio usage limits” remains an explicitly advertised paid benefit.
I don’t primarily subscribe for the Gemini app
I understand that AI Studio does not have to use exactly the same quota structure as the Gemini app.
They are different products, and different limits can be reasonable.
But a paid tier should still be meaningfully distinguishable from the non-subscribed experience.
Some of us do not primarily subscribe for the Gemini app.
I prefer AI Studio.
AI Studio is my primary way of using Gemini, and increased AI Studio usage was genuinely one of the reasons I maintained my Google AI Pro subscription.
If the practical benefit becomes:
“Almost the free quota, but with a PRO badge”
then the value proposition changes considerably for users like me.
If Google intends Pro usage in AI Studio to be only slightly higher than non-subscribed usage, that is a product decision Google is entitled to make.
But if so, users deserve a much clearer explanation of what “increased usage limits” actually means in practice.
Low absolute limits + a daily reset is a particularly bad combination
There is also the reset behavior.
In my current experience, once I hit the AI Studio limit, I am effectively done until the next daily reset.
If the practical allowance for my workload is only around 15–20 generations, that makes the restriction dramatically more severe.
Even 20 generations with a shorter multi-hour refresh would already be restrictive.
With a daily reset, the experience becomes:
“Here are today’s prompts. Try not to waste them. See you tomorrow.”
Is that really the intended experience for people doing serious iterative work in AI Studio?
Reducing the practical allowance, reducing the usefulness of older fallback models, retaining a daily reset, and still presenting increased AI Studio usage as a meaningful Pro benefit is a difficult combination to reconcile.
From the outside, it increasingly feels less like:
“Pro gives you increased AI Studio usage.”
and more like:
“Use your small included allowance, then pay per request.”
If that is the intended direction, Google can make that decision.
But then please make it clear.
Explain the practical distinction between non-subscribed, Pro, and Ultra usage so customers can make an informed decision about whether the subscription is worthwhile for their use case.
I already contacted Google support
I have already contacted Google One support about this.
The support agent was polite and helpful, and I have no complaint about them personally.
My case was escalated for specialist investigation, and I am currently waiting to learn whether what I’m seeing is expected behavior or some kind of quota / entitlement / provisioning issue.
Technical investigations can take time. I understand that.
What is frustrating is the contrast.
Customers can apparently spend days waiting to find out what their paid subscription is actually supposed to provide, while changes moving additional fallback options toward paid usage can arrive immediately.
I know customer support, AI Studio engineering, billing, and product management are probably separate teams.
But from a customer’s perspective:
Customers do not subscribe to Google’s org chart. We subscribe to the product.
Internal organizational boundaries may explain why different things happen at different speeds, but they do not change the customer experience.
If the product can change very quickly while users then have to wait days simply to find out whether the resulting limits are intentional or broken, that is frustrating — especially when increased AI Studio usage is explicitly listed as a benefit of the subscription they are paying for.
At this point, I really just want a straightforward answer:
Is this broken, or is this actually the new Google AI Pro experience?
If it is a bug, please fix it.
If it is intentional, please tell us clearly.
Either answer would be better than forcing paying users to reverse-engineer their own quotas by manually counting prompts.
This also changed extremely suddenly for me
The timing is another reason I find this difficult to dismiss.
Just before 3.8 Flash arrived, I was mainly using 3.7 Flash.
In my experience, I could use AI Studio throughout most of the day without even thinking about the quota.
I am not claiming it was literally unlimited. Obviously some limit existed.
My point is that under normal use, I rarely encountered it.
Same account.
Same subscription.
Same general long-form RP workload.
I didn’t count every generation.
I didn’t have to wonder whether one more prompt would end my AI Studio usage until tomorrow.
I could simply use the product.
Now I am literally counting prompts because roughly 15–20 generations can be enough to end my day.
Please.
I am not asking for unlimited compute.
I am asking how the experience went from:
“I barely noticed the quota existed.”
to:
“Better count every prompt, because this might be your last one until tomorrow.”
in what feels like a matter of days.
That does not feel like a minor adjustment.
From the user side, it feels like the usable allowance fell off a cliff.
And at almost the same time, some of the older fallback options are moving behind paid usage.
If all of this is intentional, please just explain what changed.
Because from the outside, it currently looks like the quota became dramatically more restrictive and paid fallback options arrived much faster than any explanation of what happened to the “increased usage limits” Pro users were paying for.
And no, this is not really about roleplay
I realize some people may not relate to my specific use case because I mainly use AI Studio for creative writing and roleplay.
That’s fair.
But whether you use AI Studio for coding, mathematics, data analysis, modeling, research, writing, or anything else, I think the underlying frustration is the same:
the usable number of generations has been absolutely crushed.
The workloads may be different, but anyone doing serious iterative work runs into the same basic problem if they have to spend the entire session counting prompts and wondering whether the next request will lock them out until tomorrow.
So even if my particular workload sounds less “serious” than coding or technical work, the core complaint is not really about roleplay.
It is about a paid tool becoming dramatically less usable, very suddenly, without a clear explanation.
That is the part I suspect many AI Studio users can relate to, regardless of what they actually use the models for.
