Can custom openai connections try and load up the model list from /models so the user doesn't have to type the model name in manually?