Replace generic http API request with Assistant for image recognition feature


I am connecting a mobile app to ChatGPT and I have a feature which sends a base64 encoded image to the gpt-4-vision-preview model. I also send a text input with instructions on what exactly the answer should be based on the image recognition.

I want to avoid sending the instructions each time to save tokens and thought I can make it via custom GPT but then I learned they cannot be connected through the API. Then I found that this is what Assistants are for but I have trouble modifying my code to successfully connect to my Assistant.

  1. Does the Assistant supports base64 encoded images as input at all and can I connect to the same gpt-4-vision-preview model? I don’t see it from the drop down menu when I create my Assistant.

  2. If this is possible do be done, what are the not so obvious advantages of using Assistant instead of a generic request with 150 symbols of instructions?