Hello.
I think I found a typo on the page https://platform.openai.com/docs/models/gpt-3-5.
It says:
gpt-3.5-turbo | Currently points to
gpt-3.5-turbo-0613.
However, a few weeks ago when I was visiting the page it said the following:
Mentioning that the default 3.5-turbo model will be pointing to 1106. So did OpenAI change the plans or the default 3.5-turbo is indeed the 1106 version?
Hi!
I did a quick test and for me, gpt-3.5-turbo shows up as GPT-3.5-turbo-0613 in billing. so it doesn’t seem to be an error.
turbo-1106 has different pricing, maybe they forgot and decided not to once they realized that.
Overall, I don’t think it’s a good idea to use the arbitrary endpoints anyways (“gpt-4”, “gpt-3.5-turbo”); it’s probably best to lock your model in.
The
are not as arbitrary as you think. The ending shows the date that they released the model, 1106 is November 6th and 0613 is June 13th.
Ah, with arbitrary I meant the literal “gpt-4” or “ gpt-3.5-turbo” designators
The reason that the model alias pointers have not been updated is because there is a significant issue with functions and their encoding of UTF-8 in the previews, needing a new model to be trained.
Also a lot of pushback about the quality of these AI models. A switch would break many production apps that require useful output and not reduced instruction-following.
Thanks guys! Now I will double-check all promises on the API page about model changes.
