I’m currently making requests to gpt-4-vision-preview, and it does a superb job at capturing the essence of old torn, weather-beaten, often handwritten documents. But I need structured output (JSON) and I have to jump through crazy prompting hoops to get correctly structured JSON out of that model, since it lacks JSON Mode. It also lacks system prompts and Function Calling.
Has anyone seen any guidance on likely arrival of a gpt-4-vision model that’s leveled up to the latest gpt-4-turbo features?