NVIDIA has published Kumo Relational weights and an Apache-2.0 supporting library; performance remains publisher-reported.
NVIDIA releases Kumo Relational for zero-training database prediction
2026-10-22Frontier watchTRIAL
NVIDIA released Kumo Relational, an open-weight model that predicts rows in relational databases from a task prompt and up to 10,000 historical examples in context. It combines row embeddings with relational message passing and supports classification and regression without task-specific training. The model uses the OpenMDW 1.1 license, while the supporting library is Apache-2.0.
What this means for you
For teams maintaining tabular ML pipelines, this could replace feature engineering and per-task retraining for some database prediction jobs. The vendor benchmark and custom model license require a controlled trial before architectural commitment.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Architecture | row embeddings plus relational message passing |
|---|---|
| Context | up to 10,000 historical examples per prediction task |
| Tasks | classification and regression without task-specific training |
| Licenses | OpenMDW 1.1 model; Apache-2.0 structured-data-models library |
| Evaluation | publisher comparison against supervised baselines; no independent reproduction found |
| Time Basis | official Hugging Face article published 2026-10-21T18:26:30Z |
Technical impact 4
Engineering value 4
Adoption signal 2
Maturity 3
Verification cost 3
Risk 3