It depends on the field. In fast moving fields like machine learning it is common to cite unreviewed stuff on arXiv. In slower fields it is less common/accepted.