hello everyone, I kindly ask you for clarification on two issues:
if I have a chatbot that is based on a gpt-4 engine, and it is “connected” with an API key to my OpenAI premium account, will I still have the message limit speech (40 messages every 3 hours) on my bot? Or will I not?
What are tiers, what do they mean, how do they work? At the moment precisely I have a premium account, but I am still tier 0, how is that possible?
I’m tier 0 because I see on the “rate limits” page that as models I only have gpt-3.5, not 4, which I would need. So I have to upgrade to tier 1 I guess or later; but how much do I pay for tier 1? Please explain it to me in simple terms, whoever you are! thank you very much in advance
Hi, you have some good questions that are not straightforward in some “welcome to API” documentation.
Your ChatGPT subscription has nothing to do with the billing or entitlements of API use;
The API is pay-per-use, with you paying for the amount of language data that is sent and generated, measured in tokens. OpenAI API Pricing;
The “tiers” originated in the middle of last year, along with the prepayment of credits in order to use the API. Before that change to new accounts, the API was billed monthly for the use in the last month (recently with some accounts that were “grandfathered” even being kicked off this type of billing);
Tiers reflect a payment trust level, how willing OpenAI is to expose themselves to risks, such as potential credit card fraud or chargebacks. Increasing tiers involves not just prepaying larger amounts, but also a waiting period before then making a second payment for a recalculation.
What was merely a payment-slowing system is now also a rate limit system, how much you can spend per month and how fast you can call on the API. Rate limits & usage tiers
When an account is created, it may be granted a $5 free trial that expires after three months. This allows you basic exploration of the API, but at low rate limits and without GPT-4 being turned on;
To move to “tier 1”, you need to purchase a prepaid credit after adding your credit card payment method and billing info. You can add $5 to unlock the service to usable levels, although tier-1 also may get slower language output from AI models.
So: prepay, and you can be sending GPT-4 not 40 queries per hour, but 40+ in minutes, hundreds per day. At the first tier. When you are paying for each call.
so i pay for tier 1 and i have no limit of 40 posts every 3 hours? Or do I have to directly upgrade to tier 2? And the payment, do I have to do it monthly or does it go automatically?
Meaning, if I pay for tier 1 tomorrow, do I have to pay back the next month to “confirm I stay in tier 1” or does it automatically scale money to my card?
Also, if tier 1 is not enough for me and I want tier 2, do I have to go through tier 1? How would it work in this case.
I have a chatbot on my site, which is visited daily by about 120k people, so I obviously need a bot that doesn’t get stuck with the 40 posts in 3 hours limit. Economically, it’s not a problem (of course, I hope it’s not billing like 5k a month).
This is a high level summary and there are per-model exceptions to these limits (e.g. some legacy models or models with larger context windows have different rate limits). To view the exact rate limits per model for your account, visit the limits section of your account settings.
The tier system is considered by the total amount you’ve paid in the past or prepaid to OpenAI, plus the waiting period of different levels to make yourself and your payments more trusted.
You don’t directly “pay for” the tier. It is just from past successful payments. There is no backsliding (yet), you only go up in tier the more you pay (you don’t actually have to use prepaid credits for any consideration, which expire in a year).
So: you purchase $25 or whatever you might use in weeks or months. You are at tier 1.
When the total you’ve paid is over $50 plus the waiting period is over, another tier escalation and you are at tier-2. Getting pretty quickly to where you can spend several dollars a minute and provide professional AI services.
Or just pay $5 after your free trial ends after doing phone verification of API key to start, and casually personally use gpt-3.5-turbo models…
okay, thank you very much! last question/doubt: so if i pay 25$ i stay at tier 1 (i don’t have limit of 40 posts every 3 hours, correct?). And for tier 2 ($50), will OpenAI automatically scale them on my card so “when I get there”, do I understand correctly?
Via the platform you can manage your personal API access. Meaning if you actually develop/program your own application using GPT-4 for example.
If you only use ChatGPT, you can most often completely ignore this part of your account.
The message limits you are referring to are only for ChatGPT.
When you use ChatGPT you don’t have to worry about the API either. Everything is handled for you.
If you want to build with OpenAI then you will get your own personal API access, via https://platform.openai.com
In short: ChatGPT is different from the API.
If you need to manage your account you are supposed to do this via chat.openai.com
hello again, I have questions about this: when user __j wrote this “Getting pretty quickly to where you can spend several dollars a minute,” what did he mean? Because I get a little “scared” of what the overall spending is; is there any way to see the billing away day by day? Or will the billing come to me at the end of the month?
In the sense, a counter that tells me “ok lorenzo, as of today 05/03 you have used xk input and xk output, for a cost of x€,” so that at least I have an account of the costs I have to incur.
It lets you switch between a “cost” view, and a “requests/tokens” view (in the second, you must also pick the models you want to see).
You can set personal monthly levels where you are either warned by email or you’ll get turned off, in the limits page.
The result on the usage page are only by the day, whereas before November 6 (and the need to hide egregious assistants endpoint costs) there was usage down to five minutes and individual API calls.
The ability to spend $10000 a month or more at higher tiers just reflects the API is where you can build high-volume customer-facing products.
but excuse me, though, if I go to the page that you kindly linked I see at the top “usage-monthly spend,” so I guess it’s monthly billing that I see and not daily. And by the way how often does the money get deducted from my card? At the end of the month? or for every API call (so continuously)?
Let me give a more illustrative example: if I have about 12000 messages on my chatbot per month, with GPT-4, approximately how much do you think I will spend? In my estimation, between $200 and $300.
if I go to the page that you kindly linked I see at the top “usage-monthly spend,” so I guess it’s monthly billing that I see and not daily. And by the way how often does the money get deducted from my card? At the end of the month? or for every API call (so continuously)?
Let me give a more illustrative example: if I have about 12000 messages on my chatbot per month, with GPT-4, approximately how much do you think I will spend? In my estimation, between $200 and $300.
If you had a single message on your chatbot then you would be able to see that the monthly usage statistics also show a daily breakdown.
If you want to forecast the cost it makes sense to send some sample messages and test the settings and all the different options you have to manage your app and the associated costs.
The information may be presented one month at a time, but if you are an account that now uses prepay credits, your usage is continuous. Continuous deducting from your account for every use of the API, continuous adding of credits whenever you want to top up the account balance.
If you have usage, the “monthly spend” is a bar graph of individual days, and when you hover over the days, you can see a pop-up breakdown of usage by model classes like gpt-3.5 or gpt-4.
The amount you pay will be directly related to the amount of input you allow users to send, the quality of conversation you preserve every call, and the model used (and then the type of customer usage the service attracts).
A “recommend a movie” chatbot that remembers five past turns of user conversation will be less expensive than an “upload your personal journal and life story and the AI will pretend to learn everything about you” chatbot.
GPT-4 should be offered as a premium, as it costs near 20x as much to operate.
If the math of estimating a chat’s token length and the cost per token and interaction is elusive, it is going to be challenging to make wise pricing business decisions.
Hi!
now the billing usage is updating for me, but I don’t end up with the prices: 30 tokens in input (3 different input messages) and 236 tokens in output (3 different responses to the 3 input messages), and it only comes out 17cent for that, how is that possible?
If you are using Assistants, the AI maintains a conversation history in a thread and feeds it back into the AI model each time. If you enable other features like file upload, the AI can make multiple iterative calls to the tool that powers it before you see a response, each again a full context input. The unseen tool specifications themselves are tokens.
If you like prices under the programmer’s control, do not use Assistants.