Google Research introduced TabFM, a foundation model for tabular data that performs classification and regression without dataset-specific training, with every prediction coming from a single forward pass. The model reframes tabular prediction as an in-context learning problem and is available now on Hugging Face and GitHub. TabFM predicts on unseen tables with no training, tuning, or feature engineering, reading the full dataset as one prompt and predicting via in-context learning. Its architecture combines TabPFN-style row/column attention with TabICL-style in-context learning, relying on alternating row and column attention, row compression, and a dedicated Transformer for in-context learning. Training used hundreds of millions of synthetic datasets from structural causal models. Google BigQuery will expose TabFM through an AI.PREDICT SQL command soon.