It’s pointless to record at 32-bit. There’s no 32-bit A/D converters (well, there is, but they are not widely available) and even the best 24-bit converters “only” produce something like 21 to 22-bits of significant data (last 2-3 bits are just random noise).
However, if you bounce/process/render your audio dozens of times in a row you may get some advantage from keeping everything in 32-bit (floating point) domain, because accumulated quatization errors of 24-bit rendering may become audible at some point if you have some highly sensitive material during low volume parts of your music.