Nums AI, a Korean startup building foundation models for tabular data, has closed a second pre-seed round of ₩2 billion, roughly $1.5 million, according to Wowtale's Nov. 4 report. The financing brings the company's cumulative venture investment to ₩6 billion, MoneyToday reported the same day.
That larger figure matters. Several funding databases list only the $1.5 million round as the company's total, which understates what Nums AI has raised. SignalBase, for example, lists $1.5 million in total funding and omits the earlier ₩4 billion, while Millionrounds records a single $1.5 million round.
Where the money came from
The company's first financing, ₩4 billion, was announced July 20 in a company blog post that named four original investors. SignalBase lists two new investors for the latest round, though no lead investor is indicated in the available reporting. One small correction for the record: the firm's name is BASS Ventures, not the "Base Ventures" that some English-language outlets have printed.
Nums AI was also selected for ₩800 million in research support under TIPS, a Korean government-backed startup program, MoneyToday and beSUCCESS reported. That sum sits outside the ₩6 billion venture total, and it is not clear when the money will be disbursed.
A note on currency: the dollar figures are each publication's own approximation. The won amounts are those given in the Korean reports.
Causilo, the first model
Causilo, the company's debut model, launched Sept. 15, followed on Sept. 19 by a technical report posted to arXiv, the open repository for research papers.
AI Times reported that Causilo ranked first in the overall, classification and regression evaluations on TabArena, a tabular-model benchmark, when it was unveiled in September. That describes the launch moment, not a current leaderboard position. The technical report itself records weaker showings as well. On grouped-regression datasets, Causilo trails conventional methods including RealMLP, TabM, LightGBM and CatBoost.
The model is available through GitHub, Hugging Face, an API, AWS Marketplace and an on-premises option. The AWS listing is a distribution channel and should not be read as a partnership. Likewise, public code and downloadable weights do not amount to unrestricted commercial open-source use.
Customers and team
MoneyToday reported that Nums AI is working on projects with InBody, the body-composition analyzer maker. Nothing in the available reporting establishes a paying customer or a production deployment.
SignalBase's Nov. 4 listing put the team at six people, though the company has not confirmed a current headcount. Its careers page lists Research Scientist, ML Engineer and open-application categories, with no stated hiring target.
A field filling with heavyweights
The competition is not small. Google Research has introduced TabFM, a zero-shot foundation model for tabular data, meaning it is meant to handle new datasets without task-specific training. NVIDIA lists Kumo-Tabular in its model catalog. SAP completed its acquisition of Prior Labs on July 16, 2026, with preliminary consideration of about €518 million, according to SAP's Q3 2026 statement.
Then there is the matter of Jensen Huang's stage. EDaily reported Oct. 23 that Causilo appeared in his Oct. 21 GTC Berlin keynote, alongside Kumo-Tabular and Prior Labs' TabPFN-3.5. The article attributed that information to Nums AI. It is not evidence of an NVIDIA investment, partnership or endorsement.
What comes next
The Nov. 4 reports did not specify how Nums AI plans to use the new capital. The more immediate test, perhaps, is whether Causilo's September benchmark showing survives as Google, NVIDIA and SAP's Prior Labs push their own tabular models into the market.
