Google Introduces TabFM, Zero-Shot Foundation Model for Enterprise Tabular Data

Google Research has introduced TabFM, a foundation model for zero-shot classification and regression on tabular data, eliminating dataset-specific training. Trained on synthetic datasets, it uses in-context learning and ranks first on the TabArena benchmark. Google released its weights under a non-commercial licence, with code on GitHub, and plans BigQuery integration via AI.PREDICT to simplify enterprise predictive analytics workflows.