How to accurately price a gpt-4 chatbot?

So the average whale (user that uses it a lot) is going to cost me like 80c to 1 dollar a day with a chat bot,

I managed to knock my prompt from 1400 down to 500 so that helped. But I’m wondering how I should limit everything.

Does anyone else have a gpt-4 chatbot pricing model that is working for them?


1 Like

I ended up giving each user a dollar amount of credit on their account that they prepay for.

Then as ChatGPT responds back I take the “input tokens” and “output tokens” token counts the API returns in the JSON and using that I calculate how much it cost for that transaction. Then using just the math for the current API pricing per token, I deduct that dollar amount from the customer’s available “credit”.


I was thinking this solution, but unlimited margin, you let them know how much however many chats would cost per month, let them set their own limit (or not), and then bill them at the end of the month (but very clearly telling people how much theyre using so no one calls you wondering about why they have a 50 dollar bill from a chat bot they spent 300 hours talking to.)

What payment processor do you use? Is the pay-as-you-go model easier through certain payment processors versus others?

Right now I’m using the stripe webhook through firebase. I have a recurring subscription setup. I am aware stripe has usage based pricing, unsure if its with the webhook or not.

1 Like

I’m planning on showing users very prominently on the web page their remaining credit in dollars whenever they’re running ChatGPT stuff, so they’re never surprised, and can purchase more credit at will.

I’m not planning to “extend credit” myself where I go in the hole until they pay.

Regarding expectations, you can provide users the “Current Average” amount 10 pages of input and output costs, just to give them an idea how cheap it is. I plan to also have an account location where they can view actual ChatGPT token counts, but only if they dig around for that.

Sorry I can’t provide details about recommended payment processors, etc.

1 Like

I wonder how many chat apps can offer a $20/month price for GPT-4 when a single user can spend $1-2 per day by calculating the required API calls. Are they all using some caching practices?

I think apps are just charging a fee to collect revenue, and then hoping they can turn a profit before they either have to go in the red (in debt), or cut off user’s access.

It seems like an extremely unsafe strategy. While I understand that many of these apps generate millions in revenue, the math doesn’t add up.

1 Like

I will tell you this.

The user tokens are .03 per 1000

bot tokens .06 per 1000 (for gpt4 8k)

However, when you end up calculating it all (for a chatbot that is) about 95% or more of your tokens end up being prompt tokens, and 5% end up being bot tokens. And with enough data you can see what your average message will look like and that’s about .032 cents per request (user prompt + system prompt + bot completion)

Yes, it is very easy for an active user to rack up a one or two dollar charge in one day - this is why you price it with a fee every month + usage. and you let your users see how much they can chat with a given amount. So for 6$ you should be able to have roughly X amount of 20-message chats, or however you want to count it.

But I have mine set up more like a cellphone service - you have a certain usage amount that you pay for every month, plus a 7 dollar charge every month.

Be transparent with the customer entirely about tokenage, how its calculated, and a calculator to see how it all breaks down (unlike some companies AHEM)…

Then just leave it up to the consumer, what’s the point of token limiting them? Its their job to keep track of their usage, if they rack up a twenty dollar bill - that’s not the companies fault.

So that’s basically how you stay profitable, you track your tokenage and mark up your tokens based on what you know your average will always lean to - and have a monthly fee on top of usage.

1 Like

there is no way that’s what apps are doing lmao. you’d be done in two months. To stay profitable you have to have the user pay for their usage there is no way around it. It could end being more than gpt-4 , but thats simply why your product has to be better at what it does in its niche field than gpt-4 , its absolutely possible to use gpt-4 to create something that is worth it for the consumer that OpenAI themselves aren’t offering.

The real fear is if OpenAI just decides to do everything themselves and then there is truly no room for small business to compete

1 Like

I know one app that charges $4/mo and then says you can ask up to 10,000 “questions” per month. So yeah what I said IS what some apps ARE doing. They’re just trying to snag some revenue and kind of “winging it”, hoping it works out.

Wasn’t trying to insult your intelligence,

More just felt what you said was unbelievable that someone would do that I mean just last week I gave you all hell on this forum trying to figure it out LMAO.

But WOW, if you’re correct. Yeah thats way too sketchy. Just winging it and not tracking the usage. You simply can’t do that.

What you can do is charge off usage and make a product thats better than what OpenAI offers and you will indeed get customers

I notice numerous offers on the App Store and Google Play. Does this mean they all have an unstable financial strategy :man_shrugging:t2:
To provide a superior alternative to OpenAI, we need to develop a competitive pricing model. Our solution could involve offering a subscription for $20 per month, which includes 50 GPT-4 messages every 3 hours.

You should figure out what your average usage is, then charge 2–3 times that per user.

Remember the goal, presumably, is to create a great product people continue to use more and more, so the more successful you are, the more your overhead will be.

Pricing at 2–3 times your costs are gives you some room to maneuver if your costs rise unexpectedly.

1 Like

Actually I mean that no matter how good a product is, the price math is negative if you put a competitive price on subscription.

Yeah but the reality is that the churn for SaaS is 2-3 months for most users.
(Well a lot of that depends on your specific industry but the point is churn is always a factor)

If someone offers an AI therapist, and that AI therapist they find is more helpful than an actual therapist, AND its better than what standalone gpt-4 can offer and it costs 5 dollars a month plus usage, and lets say they get to 20 dollars of usage in one month…

That’s 25 dollars a month for them to replace a therapist. which would cost them 100+ a session. Is that a solution for everyone? Nope. Is it a solution for a lot of people? Yes.

Especially if you taper it so you cut out time to talk to the bot throughout your week, that way you only use it one or two days a week, than your usage is only like 5-10 dollars a month.

So I think you’re being too reductionist in how you’re viewing all of this. Your competition isn’t just OpenAI or other AI apps but the systems in place that are currently there that you can undercut with AI.


Cel phone companies have it figured out:

  • Rent-seeking minimum monthly charges
  • overage and data charges

You get paid for keeping your gym open, and get paid when someone shows up to work out daily.

The former pays for the user support costs, the freeloaders with questions that cost $10 per interaction. Follows the cost of other infrastructure instances that cost by the hour.

Yes! This (and what @wycayf suggests) is exactly what I wan to do. That way, I don’t care if they use gpt-3.5-turbo-16k or gpt-4, set their limit to 3 results or 10 – that becomes their choice. But, how are you wording this pricing structure for the average user who is used to pricing like “$5 a month for 1,000 queries”? That’s my quandary: How do I explain this without it appearing overly complex?

Now at mychatbot.domain - Boosters empower you to go farther!

Some services will limit you - we let you unlock those limits!

Buy booster packs:

  • Immediate access to continue your chat session with no hourly limit.
  • boosted chat never hits our conversation length limit.
  • Only $10 for 50 extra boosts only used when needed, good all week!

(Made up pitch and payment model)

1 Like

This is what I came up with. Interested in any feedback.

Our system is specialized as it is focused on California Real Estate Law, and only contains publications from the California Department of Real Estate. Needless to say, there is a cost to operate and maintain this system, and that cost is passed along to our users via our monthly fee mechanism. These are the components of that mechanism:


Searching with the power of AI can provide huge time savings and convenience for your CA real estate information needs. The goal with our pricing is to make AI affordable at any usage level.

Behind the scenes, the AI uses computational resources called tokens to process and respond to each search. You can think of them like the minutes in a phone plan.

Each time you execute a semantic query (i.e., you ask a question conversationally), tokens are charged for both your question and the AI response. This token cost is a hard cost, and rather than hide it in an exorbitant fee structure, we simply pass it along to you to be paid by the credits you allocate each month.


We have 3 types of Large Language Models (LLMs) which comprise our Artificial Intelligence (AI) system. Each one operates at higher levels of efficiency and accuracy:

  • Bronze (lowest level)
  • Silver (intermediate level)
  • Gold (highest level)

Bronze level will give you good results. Silver level better results. Gold level the best results possible. Our pricing gives you access to all three models at any credit level. You get to decide which one is best for you by using and comparing them yourself.

Monthly Credit

We charge a flat monthly fee of $10 for access to our system. You then select a credit amount that basically pays for the tokens you use for the month. No, there is no monthly rollover of credit

Most services charge you a flat monthly fee and limit the model(s) you can use or the number of queries you can execute. Rather than try to come up with pricing schemes that guess at what your usage will be, we give all those options to you. You decide how much credit you need for each month, how many queries you need to make and which model(s) you want to use per query.

Your token usage for each model is deducted from your monthly credits at that model’s token rate. Your account record on our system always shows you how many tokens you have used within a month. Once all of your credits are used, you can no longer execute semantic searches – but you are free to execute as many keyword searches as you like.