Hi,
I obtained access to the GPT-4 8K tokens model, but I’m interested in upgrading to the 32K tokens model.
Could you please advise me on the process for gaining access to the GPT-4 32K tokens model?
2 Likes
Hello,
There’s no way to upgrade on your own. You must wait till you be granted access to the 32k model.
I don’t have access to the 32k model at the moment. As I understand from other posts on this forum, many of us don’t have access yet. It is granted to users based on the time when they applied for GPT-4 API access.
For now, you can play with the 8k model and prepare your code for the 32k model so you can use it as soon as it is available.
To be notified when you have access to 32k, you can set up a script that checks if this specific model is available to you via API and, if yes, sends you a notification email as soon as API returns a positive result.
2 Likes
Hey silly question - on the pricing page it says “prompt” vs “completion”. Is prompt here the (up to) 32K context? So to generate a 1K token response for something with 32K context would cost
32*0.06 + 1*0.12 = $2.04
?
This is not a silly question as it confusing sometimes.
Both the prompt and completion contribute to the token count but differs in the cost calculation for GPT4.
The equation for 32K context:
<prompt_1k_count> * $0.06 + <completion_1k_count> * $0.12
You charged only for the used tokens.
For instance, the cost for 1K input and 2K output for GPT4/32K model is:
1* $0.06 + 2* $0.12 = $0.3
For more details about the cost, check this page: Pricing
Thanks for this info.
It looks like GPT-4-32k is twice the price of the 8k model.
Does it seem odd that if you used the 32k model for any <8k token prompts, you’d be paying twice as much as the 8k model? My app runs prompts shorter and longer than 8k tokens (and I often won’t know how long the prompt will be when they run) which makes upgrading less financially attractive.
Nonetheless, I’m still interested in testing the 32k model. Would anyone know the best way to gain access to that?