Yup, every second. There's roughly ~30 models created per second on average.
I'm also incredibly curious about how it's implemented though. As far as I know, reddit doesn't have an API point for all new comments, do they?
reddit's primary 'new comments' endpoint let's you fetch at most the 100 newest comments. But you have to account for times when the previously 'newest' model is more than 100 comments in the past, ie. you have a gap because there were too many comments or another request took longer than expected.
The endpoints themselves are simple, but the task to make sure you don't miss any or fall behind is tricky.