Tabular data—data organised in spreadsheets in rows and columns, or in relational databases—remains the most widespread and economically most valuable data modality in the world. It underpins decision-making across nearly every sector—finance, insurance, healthcare, energy, manufacturing, logistics, and the public sector—where structured records of entities, transactions, and measurements are the norm.
This proposal aims at no less than entirely revolutionising the field of tabular data science, leading the foundation model revolution in this trillion dollar market. Its success will determine whether the German startup Prior Labs (www.priorlabs.ai), currently leading the worldwide research effort on tabular foundation models, will be able to hold its ground against large Chinese and US companies, and will be able to shape this field with European values, focussing on trustworthy methods, making causality, explainability and robust out-of-distribution generalisation first-class citizens.
Prior Labs' TabPFNv2, published in Nature magazine in January 2025 has already started to revolutionise tabular data science: in a single neural network forward pass (taking seconds), it outperforms the most popular gradient boosting libraries XGBoost, CatBoost and LightGBM, even when these are tuned for four hours. Since its release in January 2025, TabPFNv2 has already been cited by over 400 scientific papers, including over 100 use cases in fields as diverse as Healthcare and Life Sciences (over 50 use cases alone), Financial Services, Banking, and Insurance, Energy and Utilities, Manufacturing and Industrial, and the Sciences. Prior Labs also has paying customers in many of these fields.
This proposal details Prior Labs' ambitious research agenda to tackle TabPFN's remaining limitations and expand the principles behind it to neighbouring problems, including time series forecasting, multi-table reasoning in relational databases. This is a massive bet and Prior Labs will dedicate most of its highly-skilled researchers to this proposal—currently, 20, projected to be 60 by end of 2026. We plan the following advances, detailed below:
1) Scale up TabPFN to tackle arbirtary-sized data.
2) Substantially advance TabPFN extensions for counterfactual fairness and interventional reasoning.
3) Expand TabPFN to a native time series foundation model
This project is in line with the HORIZON Research and Innovation Action ELLIOT (which is funded by the European Commission with 25M Euros for staff but no compute), where Prior Labs is responsible for scaling up pretraining of tabular foundation models and time series foundation models and deriving reproducible scaling laws together with its partners from Forschungszentrum Jülich.
The project outcomes will directly enable high-impact European applications in thousands of tabular prediction problems, including trillion-dollar use cases in finance/insurance, health/biomedical, and e-commerce/platform economy.
Frank Hutter, Prior Labs GmbH, Germany