Nums AI, a developer of tabular foundation models (TFM), said on Tuesday it has unveiled its in-house model Causilo.

The company said Causilo ranked first in overall, classification and regression Elo evaluations in TabArena, a global benchmark for tabular data AI.

The company stressed that it ranked higher than TabFM released by Google Research in June, Mitra-v2 released in September by Amazon, Xiaomi-TabLDM released in September by Xiaomi, and TabPFN-3 released in May by Prior Labs, a German AI company acquired by SAP.

Tabular data is table-form data organised in rows and columns, such as sales records, transaction histories and customer information. Causilo supports performing data prediction tasks immediately without tuning or retraining the model.

The company stressed that firms do not need to develop and manage AI models for each task. Nums AI is made up of five core founders from Seoul National University and KAIST, led by Chief Executive Jae-min Yoo (유재민), a Google PhD Fellowship awardee in machine learning.

Nums AI said it has released both the Causilo code and pretrained model weights so researchers and developers can use them directly and verify performance. It applies a separate licence for commercial use.

Nums AI previously raised its first investment worth 4 billion won, led by Stonebridge Ventures with participation from SBVA, KT Investment and Base Ventures. Yoo said, "This achievement in our first year of founding shows that the expertise of our researchers and our development capabilities, focused on tabular data AI, are top-level even on the global stage," and added, "We will connect this technology to prediction and decision-making in corporate settings and grow into a world-class tabular data AI company."