If i want to calculate costs is following formula correct:
file_size * timestamp * price?
What about 1 GB for free. Is it first GB used or anything below first GB all the time?
Let me see if I can clarify what you might expect.
You are showing the cost of storage of vector stores. That is an amount simply billed per day.
The usage is free when using Assistants. It is only the text returned from the file search being placed as model input and run again - up to 16k tokens of input bill at the token rate of the model, that also then become part of the chat thread, that will cost you in terms of “usage”.
Uploading files, the storage is free. 100GB can sit in the files endpoint and you won’t get a bill.
However, when you add those files to a vector store, that is when the costs start to accrue. You are also billed again for a different vector store, even if having the same files.
The consumption is not just the extracted chunk text from files, but also, for every chunk, there is a 1k vector. So a 800+400 (overlap) chunk may be 5k, plus the addition of the vector store, plus whatever metadata might be created.
And then the tool’s usage is still free (Responses endpoint now has a per-usage charge), not billed at any rate. You just get dinged the amount shown in the price list daily for the vector store data sitting there.
You’ll have to observe in the usage page the current size of all vector stores being billed. If you list or request a vector store object, there is a usage_bytes field.
Thank you for your reply.
Apologies for late response. Your understanding is correct. Do I interpret it correctly for case when I had file for 2 hours.
Given: I have uploaded file to vector store, let’s assume file was 1GB size and after embeddings were create it showed 1.5 GB usage.
When: I have deleted file after two hours from vector store.
Then: Pricing should be following 1.5GB * time stamp (2 hours) * 0.1$ per GB / 24 hours
Separate scenario I check platform costs and for day X I have following informations:
File search: 0.11$ per day X
Vector store: 29GB Usage
My question is how is that possible? Does OpenAI have some bugs in dashboard? We use assistants api for that. By checking all files we didn’t generate such a usage with files what is confirmed by file search costs. Do I miss something?
I kind of doubt that OpenAI is going to be so generous as to pro-rate you by the minute for the storage cost.
It is a strange business model, because what actually incurs AI usage is the initial embeddings of data to vectors (then stored).
1GB of data in a vector store then just is keeping 1/4000th of the platters magnetized on some disks somewhere.
I consider the shift of expense to ongoing storage to be carefully considered as more initially appealing, and then rent-seeking. “We evaluate your cost at 00:00 UTC only” would be invitation to delete and re-create around that time for free, so no, they’re unlikely to give you the full details or minimize your expense. “Maximum size hit in a day” would be my system.
Again, pull down a listing of all the vector stores you have by API, and audit their consumption.