Google started to announce (finally!!) a release date for Gemini 1.5 Pro API on Google Studio AI (no mention for VertexAI but I guess once this one goes the other will follow shortly) and a preview of the pricing.
OFICIAL RELEASE DATE: MAY 2
Now, before we freak out, it is really interesting they are making the 1M context available for anyone(that can pay, of course). But just this move of not gatekeeping by a special request that has to be approved, i find it really cool, if you can pay the price you can use it.
Now, analyzing the price, it can be scary putting out like that, but it will work like any other LLM api we see out there, so $7/$21 is the limit of biggest cost you can reach, but your actual cost will be a calculus of how much token you used based on this value, so let’s break this down a little.
1M Input Tokens = $7
1M Output Tokens = $21
Now, let’s see if you go for a context of half of that (still bigger than other other commercial LLM out there today):
500K Input Tokens = $3.50
500K Output Tokens = $10.50
If we go to 200K, that is a window we see more today:
200K Input Tokens = $1.40
200K Output Tokens = $4.20
Or 128K, same as GPT-4 Turbo:
128K Input Tokens = $0.89
128K Output Tokens = $2.68
Now, let’s see against what we have of max avaliable window on Google today 32K:
32K Input Tokens = $0.22
32K Output Tokens = $0.67
And just for the fun of it let’s see 16K, 8K and 4K:
16K Input Tokens = $0.11
16K Output Tokens = $0.33
8K Input Tokens = $0.05
8K Output Tokens = $0.16
4K Input Tokens = $0.02
4K Output Tokens = $0.08
Table compared with the pricing of others models today:
So, we can see Google is coming into the arena as a strong competitor, no one could really test Gemini 1.5 PRO in real world scenarios but it’s expected a GPT-4 close performance for it. If this proves to be true we have it with a really lower price compared to others in many situations.
They are also making available a free limited version for testing with 50 requests per day and 2 per minute with 32 tokens.
OBS: Now for us to know the most powerful price vs performance benefit model we would need to have reposts on how Claude 3 Haiku performance against Gemini 1.5 Pro.
0 Comments