NVIDIA Introduces Kumo Relational, a Foundation Model for Relational Databases

NVIDIA has released Kumo Relational, a pre-trained relational foundation model that makes predictions directly on enterprise databases. Given a predictive query, the model gathers up to 10,000 historical examples as context and produces outcomes such as churn, demand, or fraud in a single forward pass, with no task-specific training, feature engineering, or table flattening required.

According to NVIDIA, 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 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 (MRR), extending the lead to 16% after fine-tuning. On RelBench V1 classification tasks, it outperforms gradient-boosted trees on engineered features by nearly 4 AUROC points on average. The model also sits on the accuracy-efficiency Pareto front in the RelArena-α leaderboard, reaching similar accuracy roughly 430 times faster than the highest-scoring entry.

Kumo Relational was pre-trained on synthetic relational data and adapts to new tasks through in-context learning, reasoning within and across tables while preserving foreign-key and timestamp relationships. It scales to databases with more than 500 billion rows and runs in memory, through database and warehouse connectors, or via a custom graph engine using memory-mapped I/O. The model is available under the Apache 2.0 license with weights under the OpenMDW License 1.1, and can be deployed as an NVIDIA NIM. Developers can load it through the open-source structured-data-models library or query it using the Predictive Query Language (PQL) via a NIM client.