We introduce TabDPT, a tabular foundation model capable of providing highly accurate predictions for unseen tabular datasets with no further training or hyperparameter tuning, and demonstrate scaling in both model and pre-training dataset size.