How do you determine how much to charge for a pickaxe to ensure you are at least covering your costs …
The one I am creating might be used multiple times in a day.
Thanks!
I can’t give you a direct response, but this is how I estimate my costs.
Here is a detailed step-by-step calculation process to ensure costs are covered, considering the number of tokens used, platform fees, and credit card fees.
Step 1: Calculate Cost per Token
Assume you have the following details:
• Cost per token from the API provider (e.g., OpenAI charges $0.4 - average input/output- per 1,000,000 tokens for GPT-4o-mini).
• Average number of tokens used per session/query (you can do some tests and check on the OpenAI platform what is the average)
Let’s assume the average number of tokens used per session is 2,000.
Step 2: Calculate the Base Cost
Calculate the base cost for the tokens used in one session:
Base Cost per Session = (Cost per 1,000 Tokens)/1,000 x Average Number of Tokens per Session
For example, if the cost per 1,000 tokens is $0.0004:
Base Cost per Session = 0.0004/1,000 x 2,000 = $0.0008
Step 3: Include Platform Fees
Platform fees (Pickaxe) are 10% of the final charge. Let P be the price you charge to the user. The platform fee is:
Platform Fee = 0.10 * P
Step 4: Include Credit Card Fees
Let’s estimate the credit card fees to be 5% of the final charge. The credit card fee is:
Credit Card Fee = 0.05 * P
(this is a simplified view, as credit card transactions are not applied per session, and there is a fixed part in the credit card fee)
Step 5: Total Cost Calculation
The total cost must cover the base cost, platform fees, and credit card fees.
P = Base Cost per Session + Platform Fee + Credit Card Fee
Step 6: Solve for P
P = $0.0008 + 0.15P
P - 0.15P = $0.0008
P = $0.00095 (rounded at $0.001)
Step 7: Final Price
You need to charge approximately $0.001 per session to cover your costs.
10 sessions per day
20 days per month
Minimum Monthly Subscription Price MMSP = 0.001 x 10 x 20 = $0.2
9 Likes
WOWOWOWOW!
That was an amazingly thorough answer!
Thank you so much for spending the time to compose that. I really really appreciate it!
Blessings,
Ed
1 Like
Thanks to @digitalandgrowth for the detailed write-up! It’s better than I was going to write.
I’ll just add a couple resources and comments.
API vs. Credits
First off, for anyone running a heavy usage Studio, I highly recommend you use your own API key rather than credits from Pickaxe. Our credits help you get started, and it is possible to use the credits at an advantageous cost, but generally speaking it will be more expensive to use our credits rather than your own API key.
API Usage Page
If you’re using your own API key, most model providers offer a dedicated cost/usage page that shows all your costs. For example, here is the OpenAI API usage page. It details all the costs tied to your API key, broken down by day. It provides great detail on your costs, and even differentiates between different models.
Model Costs
All models have a pricing page that breaks down the cost of using their models via API. Here is the OpenAI model cost breakdowns.
Expensive vs Inexpensive Models
It’s also important to be aware of which models are more expensive and which are cheaper. In most cases that I’ve seen, the cost of the model usage is well below the price customers are charging. However, if your tools are particularly sensitive to cost (the prompt are super long), try switching to an inexpensive model. For example, OpenAI’s newest model GPT-4o mini is very cheap. It is fractions of a cent for each use.
3 Likes
From my experience, it’s really hard to charge less than the cost of the LLM usage.
Most tools are used a few times a day at best; if you somehow create a tool that people are depending on multiple times a day then they will pay a large premium for that.
For American customers, $5 a month seems to be the base subscription fee — I almost never see anything less than that. But let’s say you went down to a $1. Provided you are using 4o mini, I think you’d have very few users logging $1 or more a month of usage on your pickaxe.
On the other hand, making a product that you aren’t losing money in terms of your labor costs is an entirely different calculation.
3 Likes
This is an amazingly helpful post- one of my many concerns (the most pressing one is the best way to sell my subscriptions) is to make sure that no matter how much my clients use the service I am always cash flow positive!
For my services I will prolly use GPT4 which my pickaxe studio says is Open AI’s best. (However given that we can choose from 9 different platform which is BEST because Pickaxe says Sonnet 3.5 is the SMARTEST?) I need to make sure that we give our subscribers the best quality output available.
Would you please be so kind as to also give the costs of using GPT4 and Sonnet?
Thanks and have a great week everyone!
2 Likes
@admin_mike thanks for this.
A) So with 9 platforms to choose from and three providers (Open AI, Mistral, Sonnet) I suspect that I will need to now get and use THREE different API’s? Are there any technical and logistical challenges to this?
B) I have also seen some platforms that allow me to set the AI choice to automatic and the platform chooses the cheapest option for the prompts reply. Maybe something to think about in the future.
C) I really appreciate that you are transparent about using the API keys are so much cheaper. Apology for the ignorant question- I imagine the reason you charge more if we buy credits is because if you were to make the credits cheaper to buy from you that would cause massive compute bandwidth on your end?
I love your service and love your business model that puts you on the same side of the table as me 
Have a fantastic week.
3 Likes
You can see all the costs for OpenAI related models here.
If you want to use all 3 models you will need 3 different API keys. Typically, different people like different models. I suggest sticking to one provider, like OpenAI and using their models. It will have the advantge of only needing one API key and you will get accustomed to the eccentricities of their models.
Automatic prompt choosing is an intriguing idea. At the current moment it has the downside of make the outputs less predictable. All the models behave a little differently. They aren’t exactly perfectly interchangeable.
The reason credits cost more is because we don’t know whether you’re gonna make a request with 10 words or 10,000 words. So we have to average the cost of credits as if you’re doing more expensive requests. Early on we had problems with people taking us for a free ride based on vulnerabilities in our cost structure. You can read more about our credit system here.
2 Likes
@digitalandgrowth Thanks for this amazing formula. We will use GPT4-o which has a higher cost. Here is my question- when a user is on our custom domain. Is it still true that every time they hit submit they are using ONE CREDIT or ONE USE? For example, if they sign on and stay signed on and come back a week later and ask more questions will that then charge them more?
2 Likes
Hey Nancy, every single use is always one credit. Whether they are using GPT 3.5 or GPT-4.
However, if you are running your tools via API key, each use will be billed exactly at cost in your API account. You are operating outside of the credit system entirely.
@admin_mike thanks for this. So now I understand what I am paying. Let me clarify- I charge clients to use my bots on a tier system. My clients who I charge, with your platform, what is the basis of what I am charging them? IE Per prompt, per session, per token etc?
1 Like
It depends on whether you are using Pickaxe credits or have it connected on each individual Pickaxe level to an API account.
If they are using Pickaxe credits then every time the user hits submit in a conversation it would be one credit, as best I understand it.
If the Studio is running on API’s then its whatever the LLM company is billing you for the usage.
With the Studio you sell uses. So each time they get a generated answer from one of your Pickaxes, that is one 1 use.
@intellibotique thank you so much sir. Have an amazing rest of the day!
@admin_mike thanks so much for your patience now I get it- U ROCK!
This is what I determines was the average (input + output)/2 for each engine…
Can someone verify to make sure I am not crazy?
Average cost per 1000 tokens
GPT-4o-mini (Fastest) = 0.000375
GPT-4o (Biggest context window) = 0.005
GPT-4-turbo = 0.02
GPT-4 = 0.045
GPT-3.5 (Cheapest) = 0.001
Mistral-Small = 0.002
Mistral-Large = 0.008
Claude-3-Opus = 0.045
Claude-3.5-Sonnet (Smartest) = 0.009
If I did it right, it seems like GPT-4o-mini is the cheapest, not GPT-3.5
1 Like
Wow this is thorough. Thank you for being so active in the community. I am so far excited that I made this purchase. I hope it really addresses the needs I’ve had for my business.
@digitalandgrowth
Hope things are well with you. So if we have understood your formula using Chatgpt mini you said “You need to charge approximately $0.001 per sessionn to cover your costs.”
Previously @admin_mike said the following.
Apology for being so obtuse and tedious @digitalandgrowth in your great formula is your SESSION the same operational definition as the @admin_mike definition of one USE?
I have created a pickaxe based on the amazing thread here!
Give it a whirl!
Would love feedback to make it better.
3 Likes
Hey Nancy, it’s a little different depending on whether you’re using Credits or own API key. Pretty much all power users who are making money are using their own API key. Here’s how it breaks down.
If you use Pickaxe Credits, every generation (whether it’s 10 words or 1000 words) is 1 credit.
If you use your own API key, then you are billed by tokens (which are roughly equivalent to words). Each model has different costs, but GPT-4o mini for example is about $0.00015 for every 1000 tokens. GPT-4o is about $0.005 for every 1000 words.
OpenAI provides a pricing chart about API usage.