NVIDIA launches Kumo Tabular, open-weight foundation models for tabular data
jure · x · 2026-09-29
NVIDIA's Jure Leskovec announced Kumo Tabular, a family of foundation models for tabular data. Like foundation models did for text and images, it lets users provide labeled tables plus rows to predict and get results in a single forward pass — no task-specific training, fine-tuning, or feature engineering.
Key points:
- Establishes a new Pareto frontier across the accuracy–inference-time tradeoff
- Fully open: open weights, open-source software, permissive commercial license
- Released on Hugging Face and GitHub
- Part of a fast-growing research ecosystem around tabular foundation models in academia and industry
Related event: NVIDIA Open-Sources Kumo Tabular Foundation Models for Tabular Data(3 posts)→
More from Research
- TCRvdb: a functionally validated TCR-pMHC database for specificity models — iskander · 2026-09-30
- Task Scheduling as a Bandit Problem: FLD Paper Details Continuous-Time Bandit in Human Motion Learning — breadli428 · 2026-09-30
- RoboPapers teases episode on steering pretrained robot policies with VLMs (VLS) — micoolcho · 2026-09-30
- VisionHOPE: First Visual Backbone Formulated as a Self-Modifying Learning System — CASIA · 2026-09-30
- Microsoft Trains a 'Night Science' Agent with RL, Expanding Research Directions 27.8% — MicrosoftResearch · 2026-09-30
- SentZero: sentence-centric VL pretraining boosts zero-shot multi-task chest X-ray analysis — Hangyul Yoon · 2026-09-30