I have been using a couple of OpenAI models such as text-embedding-ada-002 and text-davinci-003, among others, for my projects. I noticed that the pricing page on OpenAI’s official website (Pricing) only lists the prices for the “Ada” and “Davinci” models. I am not sure if these prices apply to the specific versions of these models I am using.
Is there an official price list for these specific models? If not, should I take into account the price of the general model (i.e., “Ada” for text-embedding-ada-002 and “Davinci” for text-davinci-003) when calculating the cost?
I appreciate any clarification or direction on this. Thank you in advance for your help!
The Price Model names cover the 002 part for ada and the 003 part for DaVinci, the old models are no longer supported, so the price you see if for the latest variant of each named model, note that “fine-tuned” versions of models are listed in the fine-tuned section at a different (higher) price.
For calculating costs you should use the ada pricing when you are creating embeddings and the davinchi or GPT-3.5Turbo, etc. etc. when inferencing the results from embedding retrievals.
Typically this could be:
Cost per token for ada-002 to embed your entire corpus as a one time fee.
Cost per token for ada-002 to create a vector to match against your embedding database + cost per token sent as a prompt to Davinchi-003 + cost per token sent back in reply to your prompt from the Davinchi-003.
I greatly appreciate your swift and thorough response; it was unexpected! I will take some time to go through it. Thank you for the valuable information.
I have a follow-up question. If I use the “text-davinci-003” model, I see the DaVinci model listed only under the “Fine-tuning models” in the pricing page. But since this model falls under the older models, should I refer to OpenAI Platform (Deprecations), where the price for this model is $0.0200 per 1K tokens? Is this correct? Thank you very much in advance.
If you wish to use a fine tuned model, then yes, I would not start a project with Davinci-003 though. GPT-3.5-Turbo or GPT-4 are more powerful models with greater capability, and in the case of GPT-3.5-Turbo it is also far cheaper.
I agree and disagree at the same time because I have come to the same conclusion but now I cannot access the gpt4 API on my personal account because the access is not being granted.
It’s reasonable to assume that something similar may happen with the ability to fine-tune the new models. This is what happened with the plug-in developer access, the 32k content window and ultimately depends on factors we cannot influence.