#Embedding with Davinci takes more tokens than embedding with Ada 2?I'

1 messages · Page 1 of 1 (latest)

honest siren
#

I'm embedding a very large document with Davinci (trust me, I'd love to use Ada 2 but it consistently produces worse results) and I was expecting to be able to calculate how much it would cost based on how much embedding the same document with Ada 2 cost. After embedding with Ada 2, the Usage dashboard said I had embedded 4M tokens which (at $0.02/1000 tokens) would be $80 on Davinci. But I'm running the same document through Davinci and I'm already over $600 and I have no idea how much more is left! How do I do the math on this?

#

Ok now it's done and it cost ~$622 to create embeddings for the entire document. However, if I look at the daily usage breakdown, it only shows 3M tokens run through text-search-davinci. How is that $647? What am I missing?

valid flame
honest siren
#

I think it's more like 5000x more expensive

valid flame
#

probably, i think it was deprecated 1-2 years ago and will be deleted soon

honest siren
#

yep scheduled for shutdown 2024-01-04

#

did you have thoughts on the reason the cost is so different from what I expected?

valid flame
#

no, but stats might be delayed for so old models

honest siren
#

Sure but why would the total number of tokens embedded be different for two different models? I embedded with Ada2 first and it was 4M tokens which (at $0.02/1000 tokens) is $80, not $800.

valid flame
#

ada 2 is $0.0001 / 1K tokens

honest siren
#

yep

valid flame
#

davinci was 0.01$ or something

honest siren
#

sorry Davinci is $0.02/1000 tokens

#

I was saying I ran Ada2 first to see how many tokens it would be and then ran Davinci

#

At this point, I'm actually wondering if their math is wrong or the price listed is wrong...

#

like maybe they're billing at $0.2/1k tokens evne though their pricing chart says $0.02

#

oh

valid flame
#

which chart?

honest siren
#

ugh

#

I was looking at the wrong chart

#

yeah it's $0.2/1k