NVIDIA has introduced Kumo Relational, a pre-trained foundation model designed to make predictions directly on relational databases. Instead of flattening tables and engineering features, it reads the database structure—tables, foreign keys, and timestamps—and answers predictive queries in a single forward pass. At inference time, it gathers up to 10,000 historical examples as context, using in-context learning to adapt to new tasks without training.
According to the announcement, Kumo Relational is the first few-shot foundation model to outperform supervised machine learning on common relational benchmarks. Across 41 tasks on 15 databases, it beats supervised relational deep learning models and expert-engineered gradient-boosted tree pipelines. On the SAP SALT enterprise benchmark, it surpasses AutoGluon ensembles by 8% in mean reciprocal rank, and the authors say it also beats XGBoost on engineered features by 7 MRR points out of the box.
The model is built to scale to databases with more than 500 billion rows and is available through NVIDIA's open-source, GPU-native structured-data-models library or as an NVIDIA NIM. The authors frame Kumo Relational as part of a broader shift toward structured data foundation models, where a single pre-trained model replaces task-specific pipelines and lets agents ask predictive questions about churn, demand, or fraud. The paper is available on arXiv and documentation is linked from the announcement.