cost to use gemini for embedding?

[PaxPax40]

I got around 20k notes with 4 million words, mostly smaller notes with 200-500 words each, I am adding to it progressively by scraping data from books and papers, what is the cost factor I am looking at initializing and re-embedding each time I add to it?

[brianpetro]

Gemini embeddings are currently $0.15 per million tokens.

Just a quick estimate based on your word count of 4million: I'd expect the initial embedding process to cost up to $3 with "blocks" enabled and as low as $1 with blocks turned off.

The initial embedding process should only occur once unless you delete all the embeddings by either deleting the .smart-env folder or using the "Clear sources data" button in the settings.

New notes will be embedded as they are added to the vault. But the additional per new note cost will only reflect the amount of words in the new note, so likely less than 1/20,000th the initial processing cost per note.

If cost is an issue, I do recommend sticking with the default local model which is always free to use, and the model that I use personally 🌴

[PaxPax40]

@brianpetro really appreciate you replying, cost doesn't seem like an issue yet, but how can I know to make an estimate myself? I am progressively adding to the vault and each day it grows by a dozen notes, on productive days I touch a hundred new notes made including complete chapters from books.

I am guessing for 4 million words, the estimate is 4 times 0.15$ which is 60 cents rounded off to a dollar? and with blocks enabled, 340.15 which is roughly 3$?

[brianpetro]

The estimate is very rough and depends on average word length and language. My calulculatiomn assumes 1 word ~= 1 token.

The important thing to note about blocks is that it roughly doubles the amount of tokens processed by the embedding model, but this also depends on how each note is formatted and the "minimum embedding size" setting 🌴