OpenAI's previous completion endpoint for the davinci-003 and older models included a "logprob" return option: https://platform.openai.com/docs/api-reference/co

OpenAI's previous completion endpoint for the davinci-003 and older models included a "logprob" return option: https://platform.openai.com/docs/api-reference/completions/create#completions/create-logprobs https://platform.openai.com/docs/api-reference/completions/c... Their newer chat style endpoint for the GPT-3.5-turbo and GPT-4 models no longer supports this. https://platform.openai.com/docs/api-reference/chat https://platform.openai.com/docs/api-reference/chat