Nums AI's Causilo shown in Huang's GTC Berlin 2026 keynote
NVIDIA's CEO presented the Korean startup's tabular model alongside Kumo-Tabular and Prior Labs' TabPFN-3.5 during his October 21 keynote.
Korean startup Nums AI's tabular AI model Causilo was introduced during NVIDIA CEO Jensen Huang's keynote at GTC Berlin 2026 on October 21, according to platum.kr.
Causilo appeared alongside NVIDIA's Kumo-Tabular and Germany's Prior Labs' TabPFN-3.5 as an example of a tabular foundation model. The model is designed to predict values such as sales volume or customer churn from table-formatted corporate data, including revenue records and transaction histories.
Nums AI was founded by researchers from Seoul National University and KAIST around Yoo Jae-min, a professor in SNU's computer science department, who leads the company.
Benchmark claims and funding
In September 2026, Nums AI released Causilo and published the code and trained model weights so that researchers and developers could verify the work, with a separate license covering commercial use.
According to the company, Causilo ranked first overall and in both the classification and regression categories of the TabArena benchmark at the time of its release. Nums AI previously received a first investment of 4 billion won from Stonebridge Ventures, SBVA, KT Investment and Base Ventures.
Quick answers
What is Causilo?
It is a tabular foundation model from the Korean startup Nums AI that predicts values such as sales volume or customer churn from table-formatted corporate data.
When was Causilo released?
Nums AI released the model in September 2026, publishing the code and trained model weights with a separate license for commercial use.
Who founded Nums AI?
The company was founded by researchers from Seoul National University and KAIST around Yoo Jae-min, a professor in SNU's computer science department.