Google Cloud introduces TabFM in BigQuery for zero-shot predictive analytics

TabFM, developed by Google Research, is a state-of-the-art, pre-trained foundation model for regression and classification tasks on tabular data. It leverages in-context learning to deliver highly accurate, zero-shot predictions instantly via a single SQL statement, eliminating traditional model training and deployment. Users can access TabFM through new built-in SQL functions like AI.PREDICT for instant predictions and AI.EVALUATE for performance assessment, simplifying predictive ML for various business applications.

TabFM is ideal for rapid, high-quality predictive insights without deep ML expertise, especially with small-to-medium datasets, frequently changing data, or agentic applications requiring on-demand analysis. Traditional models like XGBoost remain suitable for very large historical datasets, custom hyperparameter tuning needs, when exceeding TabFM's current feature limits, or when feature-importance explainability is crucial.