Is mixing free plan/quota and pay-as-you-go from a single app allowed and possible?

Gemini API
1

I am not sure whether this forum and/or the section is only for technological/implementation questions and mine is more of a financial/servicing question. I apologise if I posted a wrong question here.

If the costs of an AI solution needs to be desperately low (for a charitable non-profit with a very low IT budget), is it OK for a very simple web app utilising Gemini 1.0 pro (minimally fine-tuned but highly prompt engineered to stick to relevant philosophical/religious/introspection topics) to use free and paid plans together to serve two different audience and usage patterns?

Say, patient, regular, low volume, sincere users who understand the need for low costs can compromise on latency could be served through the quota limits of a free plan throughout the day but are awarded with a very rich experience with the help of very detailed prompts, acknowledging their patience with quality content.

Burst, seasonal, casual, random and “new user” traffic can be redirected to a separate pay-as-to-go subscription with highly frequent, short QA session kind interactions which will serve the volume but not be of much complexity (lower input tokens for prompts and restrictive output token).

The whole solution will still to be throttled daily to be less than say $300-$600 monthly budget initially but may increase in the future if significant value is seen to justify the budget, which has to come from the already a pool of unsolicited and totally voluntary donations. Hence, exists the need for the desperate need to keep the initial costs low.

A single web interface part of the website will be servicing both scenarios.

Will this be a) legal b) ethical c) viable?

That’s how many services like gmail, drive and many services work anyway, some even implicitly, often proving a portion of the service to be free even as a part of paid plans.

So, am I wrong to attempt to mix a free and paid plan together from the same web app? I never explicitly see any explanation anywhere regarding this, unless my searching skills are just very very bad, which they very well might be.

2

No worries, this is the right place to post this question.

A) I don’t believe it is legal but their TOS is a bit confusing.
B) I don’t believe it is ethical but if you go along with it keep in mind that Google reviewers will read all messages sent to the API via the free plan.
C) It would be viable to program it, but I am not sure about the legal implications of this.

But I don’t believe you should need to be have to do this with that budget. Here is some simple math with a budget of $300 per month (note this input is pretty high for most conversations but it also depends on what your App is for):

Input cost: (150 tokens / 1,000,000 tokens) * 7 = \0.00105
Output cost: (150 tokens / 1,000,000 tokens) * 21 = \0.00315

Total cost per conversation: 
0.00105
+
0.00105+0.00315 = $0.0042

Number of conversations: 300 / \0.0042 = approximately 71,429 conversations
3

Thanks for quick response.

I really don’t want to seem to be pushing for or pulling out an answer that I want to hear, as I am trying to drag this conversation a little longer. I am only trying to understand.

Is there a reason you don’t believe it is legal and/or ethical? is it the basic concept that free and paid should be exclusive or something like that? Would you say your opinion is something I would have seen as more common among many more?

If it “appears” that I am trying to find ways to “just not pay” (multiple emails, accounts etc), this isn’t it. This is really asking what is allowed and disallowed.

The numbers for instance is much different as I see it as I expect the interactions to be token heavy (2400+800 per exchange) like the barista bot example shown in prompt gallery.

The demand for having the model to stick to a specific spiritual context and not go crazy with all the knowledge about the world and very little knowledge of our world is essential (imaging a system you put in place in a a very conservative church to answer on behalf of the priest, when he isn’t there).

But that said, I am still wondering. BTW, it is not the end of the world if I don’t get any free quota at all and have to throttle both query volume and prompt complexity to keep costs lower.

It would be nice to go with the free plan, carefully utilise it and use paid only as much needed. And this is a non profit in India where money goes a long way towards other charitable activities and it would be my responsibility to justify the expense on this if it is any more than what is necessary.

4

I worry that Google will ban you from the services due to violating their TOS.

If a company has a free option for testing out their services and a paid version for commercial use, I don’t believe it would be ethical to use the free service, which is not intended for this purpose.

In this situation, I would recommend contacting Google and inquiring about a grant.

5

Perfect. Good To know. Thanks for clarifying.

I am reinterpreting the availability and allowance of these services this way then. - The free option is for testing (I wasn’t sure of it) only or real low volume use.

  • The moment the paid option becomes a necessity, it is assume free option isn’t applicable any more (for that particular system)

That’s fine if so.

And yes, Google grant may/will happen also if needed. I already have a high level Google exec as a possible advisory contact if I need one, who is already a donor to the org I am volunteering for.

In theory, I could ask him about this but I don’t want to because it would then seem like I am digging for more donations on the pretext of starting a new project. That’s not my job or intention and I don’t want to be misunderstood.

Once again, thanks for the response.