The limits are still measured by 3 hour intervals by conversation turns expiring after that time.
I ran up 10+. The use of GPT-4o is now not apparent from the top bar, and you have to use the model dropdown at the end of a message.
It is hard to tell if I triggered the cutoff by testing the tools, where web browsing has now been removed. Teaching it to send to tool recipients and calling python by providing my own tool descriptions now seems to have a handler to upgrade, so any GPT you can use in the future is just going to be an advertisement:
The gpt-3.5-turbo now dumb compared to past versions of API models:
