Missing presence and frequency penalty parameters in the playground

API
1

Hey, I’ve noticed that the presence and frequency penalty parameters are missing in the new playground UI but they are not marked as deprecated in the API. Is there any way I can set them in the playground as well while building my prompt?

2

You can compare the new Responses API you might be looking at (selection at upper right):

image
image290×226 3.47 KB

To that after you choose Chat Completions:

This image shows a user interface for adjusting text generation settings, including options for temperature, max tokens, stop sequences, Top P, frequency penalty, and presence penalty. (Captioned by AI)
This image shows a user interface for adjusting text generation settings, including options for temperature, max tokens, stop sequences, Top P, frequency penalty, and presence penalty. (Captioned by AI)290×436 6.15 KB

To get a larger set of sampling parameters where they still exist. Also missing is logprobs, logit_bias, audio models, o1-mini, etc. on the new endpoint.

Bigger concern could be no ability to send max_completions_tokens to a reasoning model - like if you wanted to cut off your o1-pro “high” output spending to $60 per call (100k tokens).

The image shows a menu interface with options for "Text format," "Reasoning effort," and "Store logs," with dropdown selections and a checkbox. (Captioned by AI)
The image shows a menu interface with options for "Text format," "Reasoning effort," and "Store logs," with dropdown selections and a checkbox. (Captioned by AI)573×179 8.4 KB