Large Tabular Models Excel Where LLMs Fail
TL;DR
LLMs such as ChatGPT, Claude, and Gemini struggle with larger tables because they model token sequences. Tables behave differently: row and column order can change while the underlying meaning stays the same. Startup Fundamental is pitching Large Tabular Models as the fix. Its NEXUS model launched on 5 February 2026 with US $275 million in funding and is built for predictions on structured data.
Nauti's Take
This is one of the few AI trends where the boring use case may be the valuable one. Nobody enjoys spending ten weeks on feature engineering for fraud detection, churn, or inventory forecasts just to get a model to behave.
LTMs sound less glamorous than chatbots, but they may sit much closer to real business value. The PR layer around NEXUS is obvious.
Still, the category is worth watching because company tables often matter more than any chat history.
Briefingshow
Most business data does not live in polished documents. It sits in spreadsheets, logs, CRM exports, and measurement series. If LTMs deliver on the promise from Fundamental and others, some data analysis moves from months-long custom ML projects into faster prediction systems.
The hard part remains training data, privacy, and making models work across very different domains.
Sources