For me the holy grail of existing models would be API access to the raw pretrained GPT-4 model. It would only be available in completion mode (not instruct or chat mode) but that's OK with me.
I use 3.5 && 4 equally. Limitations prevail on 4, so it forces you to use both if you are a heavy user I think.
How do you access GPT-4 through ChatGPT? Or how can you tell which version you're using?
you have to subscribed to ChatGPT plus. When you make a new chat, you have a dropdown to select the model.