I've had some fun with Suno 5, but the songs absolutely don't replace well-made human music. They're much more formulaic and over-produced. They're usually forgettable. People I play them for can always tell they're AI produced.

True music enthusiasts will holdout for a while but I think AI music will easily replace most Pop currently on the radio and streaming for your average Joe. That stuff has been "fake" as early as the mid 2000's by being quantized straight to the grid, pitched, with programmed drums, guitar, even vocals and then churned out like widgets on a conveyor belt.

Currently for me, in the type of music I have enjoyed from v4.5+. V5 of their model is a regression.

I was very impressed with v4.5+ that I could get quite good songs evocative of Devo, Yeah Yeah Yeah's, Metric, etc.

Version 5 is currently harder (or I haven't figured out a way) to generate this kind of chopped/produced sound. It doesn't follow complex style definitions and tends to generate songs that are too slow and "smoothed" over.

I find it depends how you generate it. Asking Suno to make covers of uploaded recordings tends to give much, much better results than asking it to cook a song from scratch. There are still quite a few tells that it's AI-made but it's not bad at all, at least in my experience so far.