Large Tabular Models Excel Where LLMs Fail
TL;DR
LLMs such as ChatGPT, Claude, and Gemini are strong at text, code, and slide generation, but larger tables expose a weak spot. Row order, column meaning, and non-sequential relationships do not fit cleanly into next-token prediction. Fundamental pitches NEXUS as a Large Tabular Model built for structured data. The startup emerged from stealth on 5 February 2026 with $275 million in funding and focuses on prediction over spreadsheets rather than chat output.
Nauti's Take
This is one of the more useful AI directions because it targets where companies keep their money and risk. Still, the LTM story should not be mistaken for magic: tables are messy, context is missing, column names lie, and governance remains work.
NEXUS sounds relevant and AWS is a strong trust signal, but the category still has to prove it beats well-maintained classical models plus competent analysts in real workflows.
Briefingshow
Most business problems are not hidden in elegant prompts, but in tables: revenue, risk, customer records, lab values, and transactions. If LTMs deliver, some data-science work could shift from months of feature engineering toward faster modeling on existing datasets. The real test is reliability, auditability, and whether privacy claims survive enterprise scrutiny.
Sources