Large Tabular Models Excel Where LLMs Fail
TL;DR
LLMs predict sequential text well, but larger spreadsheets expose the gap: rows and columns can move while the underlying meaning stays intact. That non-linear structure clashes with next-token modeling. Fundamental is pitching NEXUS as a Large Tabular Model for structured data. The startup emerged from stealth on 5 February 2026 with US $275 million in funding.
Nauti's Take
The hype needs some distance: part of this reads like category-building around a well-funded startup. The underlying point is real.
Business data lives in tables, and LLMs often treat tables like awkward text. If LTMs handle column logic, statistics, and sensitive data more reliably, this becomes less of a chatbot feature and more like decision infrastructure.
Briefingshow
For companies, the most valuable AI use cases often sit outside chat: transaction logs, CRM tables, marketing metrics, clinical measurements. If LTMs reduce feature engineering and bespoke model work, prediction, fraud detection, and segmentation can move closer to the teams that own the data.
Sources