Large Tabular Models Excel Where LLMs Fail
TL;DR
IEEE Spectrum frames Large Tabular Models as a new model class for structured tables, an area where general LLMs still struggle badly. Fundamental launched NEXUS after emerging from stealth in February 2026 with 275 million dollars in funding, and the model is now available through Amazon SageMaker. LTMs aim to model values, column meaning, categories and statistical relationships directly instead of treating a spreadsheet like a text sequence.
Nauti's Take
This is one of the more sensible AI waves because it targets a real gap: LLMs look clever in chat, but become vague fast when tables get serious. Still, the NEXUS story is vendor-heavy, and independent benchmarks need to prove whether the leap is large enough.
For companies, the direction makes sense anyway: a text model plus a tabular model is more plausible than a chatbot pretending every CSV is just a long sentence.
Briefingshow
The point is not just better spreadsheet handling, but a different route into enterprise data. Many critical decisions live in transactions, metrics, clinical measurements or inventories, not in prose. If LTMs prove reliable, they could automate part of the expensive, case-by-case data science work companies currently need.
Sources