To keep AI Studio fast and available for everyone, we’re adding daily usage limits for each model in the AI Studio playground, starting today. Limits are much higher for Google AI Pro and Ultra subscribers.
What this means for you
- Most people won’t notice any change.
- If you use a model very heavily in a single day, you may hit its daily limit.
- Limits are per model, so you can switch to another Gemini model and keep working.
- For sustained or high-volume work, the Gemini API
with your own API key is the best option. If you’re a Free user, upgrade to a Pro or Ultra plan.
We’ll keep watching how these limits affect real usage and adjust them as needed. If you think you’re hitting a limit unexpectedly, reply here with the model and roughly what you were doing
By the way, as far as I remember, there used to be a daily limit on the number of requests to the model (I even hit that limit sometimes). How is the new system fundamentally different from what we had before?
One more question. In this forum thread (AI Pro Subscription), there are complaints that the models quickly hit their usage limit in long chats (and the limit resets at midnight Pacific Time), but continue to work in new and short chats. Was this a bug, or was it a test of what you described in your announcement?
It appears that the behavior described in the complaints in that forum thread is indeed related to the limit you announced.
Today I encountered an issue where Gemini 3.1 Pro, in long chats with extensive context (over 300,000 tokens), generated only 5 messages for me before starting to return a quota-exceeded error. And this error occurs in all long chats with extensive context. In short chats, Gemini 3.1 Pro generates responses normally.
If that’s really the case, and it was done on purpose, then 5 messages to Gemini 3.1 Pro in long conversations with a lot of context is definitely not a “much higher limit”, @Mustan_lokhand
More details about my experience today, September 29, 2026
Volume of the conversation in which the issue occurred: approximately 830,000 tokens
What I was doing: creative writing
Account subscription status: Google AI Pro
Model: Gemini 3.1 Pro
Issue: I was only able to get 5 responses from the model, 2 of which resulted in an “Internal Error.” After that, the model hit the limit in all long chats with extensive context, but continues to work in short and new chats
If this is an intentional change on Google’s part, it feels more like paying for nothing rather than a “much higher limit”
It seems this new “daily usage limits” rollout has just reintroduced the exact same subscription entitlement bug we spent all of September fighting.
Around September 23rd, the previous backend issues actually resolved. My Pro account was correctly recognized, and I was able to run 70+ heavy context prompts a day without hitting any rate limit walls. We had exactly six days of stability.
Today, the moment this new update went live, my account immediately regressed. Despite this announcement stating that limits are “much higher for Google AI Pro and Ultra subscribers,” my paid account is once again being hard-locked after fewer than 10 prompts.
It looks like whatever code was pushed today to enforce these new daily limits has severed the backend subscription verification again, dumping paid users straight back into the unverified Free-tier bucket.
To answer the prompt on what we are doing, I am running standard daily professional workflows on 3.1 pro and getting locked out in single digits despite paying for a premium subscription.
Please stop pushing quota updates that destabilize our paid tier verifications. We just need the stable access we had on the 23rd to remain intact so we can actually use the service we are paying for.
Same question here too.
If you gave us any even slightly useful metrics for users to plan around rather than the useless ‘high, higher, highest’.
You did it with Gemini App, you already have it set in place for Dashboard to be altered. Adding a Pro, Ultra x5 and Ultra x20 and what have you set to the Free Tier, Tier 1, Tier 2 and Tier 3 seems reasonable. Or having the Free Tier recognize Pro and Ultra sub keys and adjust said Tier to match those subscriptions rather than base Free Tier.
Can we at least know if its by RPD, TPM anything would be welcome rather than just guessing and trial and error.
Seems like a very basic feature almost every other competitor has, and is generally required for effective use in any system that has these kinds of limits.
(Edit)
Beyond all of this in regards to Australian Consumer Law, New Zealand Consumer Garuntees Act and many more, there is also the concern that what you have done unilaterally alters the material value of the service without reasonable notice to customers as required by consumer contract law.
What’s more there is also the concern this has the potential to violate the principle of good faith in commercial dealings. You cannot legally hold a consumer to a limit that you have not reasonably defined. Along side the potential fact of lack of due care and skill in the application of this system.
So to respect this can the Google Team please provide the previous limits and the current limits as well as the exact time in which users where notified to at least attempt to comply with said contract law. Given that this is not an experimental service for multiple models and does not fully fall under As-Is when monetary value is applied. (Edit)
By the way, here’s a small observation from the time when this issue with limits was occurring prior to September 23, 2026. This observation remains relevant even now, after September 29, 2026.
When I hit the model’s limit in a long chat with a large number of tokens (500,000 or more), I tried the following to check whether this was a daily token limit:
I created a new empty chat
Uploaded a large PDF (about 500,000 tokens) or video (about 600,000 tokens)
Asked the model to summarize the PDF or video
The model responded normally
From this, I draw two possible conclusions:
If tokens in videos and PDFs are processed differently than tokens we enter into the input field, then the limit applies only to tokens entered by the user manually, which explains why the limit is exceeded specifically in long chats with a lot of context
If tokens in videos and PDFs are processed the same way as the tokens we enter, then perhaps the processing of long chats specifically is failing somewhere
But this doesn’t explain one fact: based on Google’s announcement from September 29, 2026, we can conclude that if the model usage limit is reached, it will be impossible to use the model at all for the remainder of the day. However, even when the limit is reached in long chats, the model can still be used in new chats, which gives hope that this is indeed a bug and not a feature.
That has been taken the wrong way. It just means each model has it’s own limits. It does not mean switching from one to another starts you at zero usage again.
