Idea Summary
Parse OpenAI chat response when creating a REST Data Source with v1/chat/completions endpoint.

Use Case
I noticed this when I was trying to upgrade to a new model (APEX 24.1). The newer models are all(?) chat models (they return an error when trying to use v1/completions instead of v1/chat/completions).

OpenAI responses are parsed correctly when creating a REST Data Source with the legacy v1/completions endpoint. E.g. using this old model

{"prompt": "#PROMPT#", "model": "gpt-3.5-turbo-instruct", "temperature": 0.7, "max_tokens": 200}

which returns

{
 "id" : "cmpl-.....",
 "object" : "text_completion",
 "created" : 1721931469,
 "model" : "gpt-3.5-turbo-instruct",
 "choices" :
 [
   {
     "text" : "\n\nOracle APEX (Application Express) is a web-based development platform for building data-driven applications. etc. ",
     "index" : 0,
     "logprobs" : null,
     "finish_reason" : "stop"
   }
 ],
 "usage" :
 {
   "prompt_tokens" : 4,
   "completion_tokens" : 118,
   "total_tokens" : 122
 }
}

The REST Data Source automatically retrieves “text” for the Parsed Data.

The new v1/chat/completions return the answer in “content”:

{
 "id" : "chatcmpl-....",
 "object" : "chat.completion",
 "created" : 1721931275,
 "model" : "gpt-4o-mini-2024-07-18",
 "choices" :
 [
   {
     "index" : 0,
     "message" :
     {
       "role" : "assistant",
       "content" : "Oracle APEX (Application Express) is a low-code development platform provided by Oracle Corporation that enables developers to build scalable, secure enterprise applications with minimal coding effort. …."
     },
     "logprobs" : null,
     "finish_reason" : "stop"
   }
 ],
 "usage" :
 {
   "prompt_tokens" : 21,
   "completion_tokens" : 425,
   "total_tokens" : 446
 },
 "system_fingerprint" : "fp_...."
}

Preferred Solution (Optional)
Retrieve either text or content as the Parsed Data.