Blender is really an amazing case study of open source software. Apart from the Linux kernel and web browsers/tools, it is perhaps the only open source software that managed to beat all the commercial software in its niche. It has rendered Maya nearly obsolete.
Meanwhile, in other niches, Microsoft Office still beats open source office suites like LibreOffice; Photoshop isn't about to give up its crown to GIMP; Lightroom isn't losing to Darktable; and FreeCAD isn't even in the rear view mirror of Solidworks.
I wonder what will be the next category of open source to pull ahead? Godot is rapidly gaining users/mindshare while Unity seems to be collapsing, but Unreal is still the king of game engines for now. Krita is a viable alternative for digital painting.
MuseScore is good enough that I haven't bothered to check back with commercial vendors. I'm pretty novice with it, however, so perhaps Sibelius power users will disagree.
I wonder if you could assemble a big corpus of individual solo instruments, then permute a cacophonous mix of them. IIRC the main training dataset is comprised of a limited number of real songs. But I think a model trained on real songs might struggle with more "out there" harmonies and mixes.