AI NexusAI Nexus
Model release

NVIDIA releases Kumo Tabular: an open tabular foundation model that needs no training

Event time 2 independent sourcesEditorial score 62/100Updated here

The short version

NVIDIA released Kumo Tabular, an open foundation model for tabular classification and regression that predicts in a single forward pass, with no training, tuning or feature engineering. It comes in three sizes from 28M to 215M parameters under the OpenMDW-1.1 license, which permits commercial use.

What changed

Tabular prediction such as churn, default, demand or price usually means building a model per dataset. Kumo Tabular turns it into one call to a pretrained model, with weights on Hugging Face and code on GitHub.

What it means for you

Data science and backend teams can use it as a baseline against existing gradient-boosting setups. The leaderboard result and the 17x speed figure are NVIDIA's own evaluation, so validate on your data.

What it is for: classification and regression on tables such as customer records, transactions, sensor logs, claims and orders.

Caveat: first place on TabArena (ELO 1950) and running 17 times faster than LimiX-2 are NVIDIA's self-reported results under a uniform single-GPU setup and have not been independently reproduced. The event time follows the post's date.

Fact check

  • VerifiedNVIDIA describes Kumo Tabular as an open foundation model for tabular classification and regression that predicts labels from tables in a single forward pass, with no training, tuning or feature engineeringEvidence
  • VerifiedIt comes in three sizes (28M to 215M parameters) and is released under the OpenMDW-1.1 license for commercial use, with code on GitHub and weights on Hugging Face at nvidia/Kumo-TabularEvidence
  • VerifiedNVIDIA says that under a uniform single RTX 6000 Pro evaluation setup, Kumo Tabular ranks first overall on TabArena with an ELO of 1950 while running 17 times faster than LimiX-2NVIDIA's own evaluation; not independently reproduced.Evidence
  • VerifiedThe post is dated September 29, 2026; the site uses the date onlyEvidence

Coverage timeline

  1. Primary sourceNVIDIA (Hugging Face)
    NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction
  2. Primary sourceNVIDIA (model card)
    nvidia/Kumo-Tabular model weights