Because of its size, for example a protein can not be modelled from the quantum chemical basis in a reasonable amount ot time without approximations. A hydrogen atom though can, manually even. Where is the boundary between those, where approximations become necessary for the results to be sensible while taking a managable amount of time for the calculations to converge?
I suppose this question is dependent on parameters, such as what computer power is available and what is considered tolerable amount of time for the calculations. Let's say one year computation on Piz Daint, but feel free to choose the parameters to your liking.