there's still a long road to commercial applications but today's hardware is simulating quantum systems beyond the scale of classical methods, for example [1]; an interesting line of work opposite to this can be found in those who improve classical methods towards such examples [2], but these are only developed because of the existing quantum hardware
Really though, today's IBM hardware is good fun to play with, eg for generating moderately large GHZ states
yes ... especially if you want to execute quantum-circuits which use a lot of qubits.
why!? one approach of the simulation of quantum-computers rely on the so called "state vector" of the machine, and its memory-usage grows exponentially.
No. They're a decent playground for prototyping hybrid algorithms but even that is limited. No one has yet published a hybrid algorithm on a rentable QC provider that has better benchmark performance than a modern CPU/GPU implementation.
Really though, today's IBM hardware is good fun to play with, eg for generating moderately large GHZ states
[1] https://arxiv.org/abs/2510.26845 [2]https://arxiv.org/abs/2306.14887