Nums AI launches Causilo, a tabular foundation model that tops TabArena benchmarks
Nums AI released Causilo, a pretrained tabular foundation model for classification and regression, with a scikit‑learn‑style API, Apache‑2.0 code, and weights hosted on Hugging Face. It employs in‑context learning: the fit call stores training rows as context and predictions are made in a single forward pass without updating the pretrained weights.
Key points
- Causilo 1.0.1 tops TabArena Elo (1794) among single tabular models, beating TabFM and EXAONE Tabular.
- Uses in‑context learning: fit stores training rows as context, predicts queries in one forward pass without weight updates.
- Runs fastest on an H100 GPU for both fit and predict; commercial deployment needs a separate license from Nums AI.
In the TabArena benchmark covering 51 datasets, Causilo achieved the highest overall Elo (1794) among single models, surpassing Google Research’s TabFM and LG AI Research’s EXAONE Tabular. A re‑run that includes system ensembles placed it third of 88. On ScoringBench it also ranked first for CRPS, R² and RMSE across 101 regression datasets. Speed tests on an H100 GPU showed it fastest on both fit and predict, though TabPFN‑3 uses less memory.
Version 1.0.1 adds median and quantile outputs, and the model runs on CUDA or CPU with Python 3.10‑3.12 and PyTorch 2.13+. Commercial use requires a separate license; the released code and weights are for research only.
Nums AI Releases Causilo: A Tabular Foundation Model That Tops TabArena Among Single Models
MarkTechPost · 16 September 2026
Loading the full article…
This text was published by MarkTechPost and written by Michal Sutter. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Introducing Mistral Large 4 · 7 src
- OpenAI boosts GPT-6 speed by 50% in 28 Days update · 1 src
- Reflection AI debuts Beam, a 501B open-weight model rivaling GLM 5.2 with 3-4x less compute · 10 src
- Author tests four AI assistants on flawed forecasting scenarios · 1 src
- Anthropic prompts users to share voice data to enhance AI models · 8 src
Comments
via GitHub Discussions