Do they really think we wouldn't notice a 92% free tier quota?

Gemini API
,
1

I honestly don’t know what’s going on at Google/DeepMind right now, but this latest move is absolute amateur hour.

I logged in today to find that the Free Tier limit for Gemini Flash has been slashed from 250 RPD (Requests Per Day) down to 20 RPD.

Let that sink in. 20.

That is a >10x reduction overnight. No blog post. No email warning. No “heads up, we’re changing the terms.” Just a silent rug-pull that effectively bricks anyone currently developing or testing on the free tier.

The “Bait and Switch” is real It’s obvious what they are doing. They got us hooked with a usable limit, and now that we’ve integrated it, they strangle the quota to force us into the paid tier. I get that “free” doesn’t last forever, but a 92% reduction without notice isn’t a policy change—it’s hostile user experience.

And don’t get me started on the paid limit pull of system (it doesn’t exist, they just send you a notice by mail…)

23 Likes
Why receiving HTTP 429 Too Many Requests
Confusion Regarding Rate Limits
Why Am I Getting a 429 Error From Gemini API?
429 error, but google team doesn't check at least
2

If I had to guess, I’d say they’re finally done milking us for data. We all know the deal: on the Free Tier, we pay with our data so they can train their models. But now that Gemini 3 is out for Pro and just around the corner for Flash, they clearly had everything they need.

3 Likes
3

If Gemini 3 flops, they might try to backtrack or if Openai does something better. Like you said, it’s just desperate bait and switch. The thing is, most people who uses Gemini for free won’t ever buy Gemini 3. It’s too expensive and some people said it’s not even that good. And 20 RPD is just dumb. Most people will look elsewhere. Plus the lack of transparency, it’s just appaling. And if you complain, they’ll always defer to ther joker, which is the TOS. But let’s be real, the TOS is just so they can do whatever they want and justify that they can remove the Free tier as they see fit. But if you do that, you might as well just remove Free forever and stop baiting people and then drop them.

6 Likes
4

Hey! We did turn down the free tier for some models in order to free up compute to focus on the continued and growing demand for Gemini 3 Pro and Nano Banana Pro.

The free tier was designed to give devs a place to quickly test the model and then switch over to the paid API once they know if Gemini is right for them, though we offer this as best effort and in some cases it is not possible to offer the free tier. Overall, you should look at the free tier, if there is one, as something we provide best effort but it is highly unstable and not likely to persist. The compute situation with how much demand there is for Gemini makes the option space extremely complex, so apologies for any hassle on this.

4 Likes
7

12.5 times reduction? Without any notice? Lower api usage isn’t the issue here. The issue is that it was without any prior notice or even a notice at all! And 12.5 times at a time? With no prior warning to migrate to paid plan (or likely in most of cases migrate to another provider). After that of course people will be furious. Even if you will revert the changes without the proper explanation I’m pretty sure that people will start migrating from the gemini api to the alternatives as this is unacceptable.

It’s more like all models

8

Yeah that is fair, we generally try to give folks updates when possible but in many cases need to act quickly. Will see what we can do to provide more heads up next time!

6 Likes
9

Why did you have to act quickly with this action? Is it like you didn’t have enough capacity (when I’m pretty sure it scales easily)? Why didn’t you send an email notifying the users even after the fact (and even right now)? If you needed to direct computing to other tasks didn’t you expect the users to move to paid plans and still using computing power by that (and as I can see by the forum people moved to paid plans)?

What is stopping you from providing more heads up this time?

1 Like
10

Are you looking for new features :rofl:

12

This is literally No free lunch theorem of google :rofl:

14

:melting_face::melting_face::face_with_steam_from_nose: you could have posted like you posted when you did Lauch free tier telling its free and this much req per day

16

It’s over guys, no more Gemini API.

3 Likes
18

Looks like Google decided to just abandon this post and do nothing about the issue.

1 Like
19

So how am I going to develop my app with such low quota ? I plan tu use paid tier after production release, but right now giving away possibility to use gemini 2.5 glash fo developers is a HUGE bang. 250 RPD was very low for testing more advanced apps but 10 ?? It’s a joke.

21

Unpopular opinion: I appreciate this change.

I’ve struggled with degradation of the LLM’s when transcribing audio. More users = more compute = worse quality.

This change means that the results while be more stable if you are willing to pay some money. I think the pricing is still very cheap and the free tier seems still decently generous for normal use.

The only thing they’ve could have done better is notifying the users beforehand.

I hope they can continue to provide a stable solution as degradation is in my opinion the worst enemy.

If you want cheap LLM’s there are plenty open source providers out there. They are just of less quality.

No, I’m not affiliated to Google. Just an entrepreneur who likes to have stable solutions for it’s clients.

1 Like
22

I think it wasn’t just “Increasing performance for paid users“. It was more like a long term plan of trying to make people use their api (which is not compatible with openai SDK, which almost everyone uses) by providing good free tier limits, so the users are locked into gemini ecosystem and then suddenly lower the limits so the users find out that their projects are not working and to quickly resolve the issue start paying. Otherwise they would at least notice beforehand. They didn’t. As they knew that the switch to another provider will take a lot of time (lost users and revenue) and money. It is not a really good move. I personally already moved to another AI provider and encourage others to do the same as this is absolutely unacceptable. Actually gemini is more suitable to my task than another option (and it is also only paid), but I don’t want to support this type of behavior.

23

Yes, the reduction is so extreme. No image gen models now, 10 or 20 text RPD, 10 TTS RPD… Massive reduction, no warning, and you need to dig into Reddit or the forums to find a google dev respond with an explanation; no public warning - just an ‘oh well, pay us’ attitude. Does not make people like me who are dabbling in developing using Gemini API want to use their service…

24

The worst part is that even if you switch to the paid tier the API is currently unusable. It keeps throwing 429 errors on all models and support’s response was basically “Upgrade to tier 3 :smile: .“ So don’t think that paying will make it any better.

3 Likes
25

If the API is not stable I am going to have a serious problem. I was just looking into the Antigravity IDE and Gemini (yes, I asked Gemini about a Google product – it does not seem to lie) told me that it worked with OpenAI and something else. It specifically said that Google developed it with more than Gemini in mind because they thought no one would use it if they were locked in. So, assuming Gemini hasn’t been programmed to flat out lie, Google is not trying to trap anyone, they know people want to preserve their options. I would never write something for one AI, because a few months from now, maybe something better comes out. As long as the switch is simple, that’s fine. Gemini should refactor the code for you anyway.

But, I’m no Google fanboy, I just really think that Gemini Pro is the best out there. I will use the best one, whoever makes it, as long as I can change when the competition comes out with something better.

Keep in mind that ChatGPT 5.2 came out one day ago. Coincidence? I don’t think so. The articles I’ve read said it’s “all hands on deck” to make sure that Gemini still out-performs ChatGPT. That could very well be what they needed the compute power for.

26

@James3 The switch was not simple. It would be simple if I just needed to change url and api key in openai library, which everyone uses. In situation when you just switch to a better model your project doesn’t become bricked while you are writing code and you are not really time limited. What google did can’t be compared to just switching providers for a better model.

27

I understand. And I would be quite unhappy if that happened. I’m really sorry that happened to you.

I was more pointing out that we know it can happen (and if we didn’t before, we sure do now), so to protect ourselves, we should write the code so it IS easy to switch.

Sometimes that’s probably not possible. What if you are using a feature that only one AI is good at? What if the prompt structure is so different that it requires major changes? Etc, etc.

So yeah, them doing it without notice was seriously unprofessional.

These companies have so much money and power it’s crazy. I’ve thought for a long time that they should be regulated. But, it doesn’t look like it’s going to happen.

I don’t know what the solution is, but we are getting into dangerous territory here. AI’s threatening people who say they are going to shut them down, AI’s who encouraged someone to commit suicide… Many people think we are 2-5 years from AGI (which I assume is minutes to days away from superhuman AGI).

And the current administration thinks the solution is zero regulation?

By the way, the same problem exists with nanotechnology, it’s just not as mature yet. I’m an expert in nanotech, and the possibilities are wonderful and also scary as Hell (just like AI).

I’ve considered starting a DAO or some other company to get ahead of these types of issues in nanotech. Massive funding would be needed (but a lot less than is being poured into AI).

If anyone is interested in such a venture, let me know. I haven’t done anything yet because most people think it’s still decades away and I haven’t gotten any traction with VC’s I’ve contacted.

But, I can tell you for a fact, because proof of concept has been done in the lab, that things like cell-sized robots that can cure any disease, computers that make today’s supercomputers look like toys, and mass production of materials that make steel look like tissue paper, are not as far away as people think. Farther than AGI, but that’s going to be our next big concern (and blessing).

Let’s hope our AI overlords let us hang around to see it.