Don’t blame the engineers for these tools. A GTO solver demands so much resources, the algorithm used in these software have to have a lot of shortcuts to make the tools “working” possible.
Actually every solver author out there defines quite well what kind of "shortcuts" are taken. In HU solvers it's usually just betting abstraction (not every bet size is included in the model). For multiway ones you still need to take heavier shortcuts.
The thing is we can often prove that solutions available today are quite close to the ultimate solution. For some more complicated cases we have tools to give us a lot of confidence those are very close to solutions without abstractions even if we don't have a formal proof yet. It was never the case with Deep Blue, Stockfish or Alpha Zero in chess and no one claims those play to ultimately correct chess. That's the big difference which you again didn't grasp.