NVIDIA Kumo Tabular is an open foundation model for tabular classification and regression, released by NVIDIA on Hugging Face. Given a labeled table, it predicts new rows in a single forward pass, no training, tuning or feature engineering. Comes in three sizes, 28M to 215M parameters, pretrained only on artificial data. Ranks first on four benchmarks: TabArena, BeyondArena, TALENT and ScoringBench.