Kumo Tabular TRIAL

Kumo Tabular is an open-weight tabular foundation model that reads labeled rows as context and classifies or regresses in one forward pass. All four leaderboard wins are vendor-run, and the training recipe and synthetic-data generator are not yet public.

Next stepCompare the Medium model with tuned tree ensembles on held-out tables for accuracy, calibration, latency and GPU cost.

Judgment reviewed 2026-10-06 · Basis: NVIDIA releases Kumo Tabular for zero-training table prediction