TabFM Outperforms XGBoost on 38 Classification and 13 Regression Benchmarks Using a Single Forward Pass; and Is Coming to BigQuery via AI.PREDICT
According to Google Research, Google has introduced TabFM; a zero-shot foundation model for tabular data classification and regression, eliminating the need for manual model training, hyperparameter tuning, and complex feature engineering that has defined the XGBoost and random forest era of enterprise machine learning.
Tabular data powers most critical business prediction tasks; from customer churn forecasting to financial fraud detection. Until now, deploying a reliable model on any new dataset required data scientists to invest countless hours in dataset-specific feature crafting and hyperparameter optimisation just to extract usable signal.
TabFM reframes tabular prediction as an in-context learning problem. Rather than fitting a model to each new dataset, it takes the entire table as a single unified prompt and generates predictions in one forward pass; just as LLMs answer questions without retraining. The model was trained entirely on hundreds of millions of synthetic datasets generated using structural causal models.
Benchmarked on TabArena across 51 datasets, TabFM consistently outperforms heavily tuned industry-standard supervised algorithms. TabFM is now live on Hugging Face and GitHub, and Google is integrating it into BigQuery; users will be able to run advanced regression and classification using a simple AI.PREDICT SQL command, no ML expertise required.
The last decade belonged to gradient boosting. If TabFM generalises as well at scale as the benchmarks suggest, the next decade of enterprise ML looks very different.
Check out our previous coverage of AI research and machine learning on The Trusted Times.
