Tabular Foundation Models

Yes. TabPFN can be deployed in secure environments, including on-premises and private cloud setups, ensuring your sensitive enterprise data never leaves your control and compliance with data protection and governance requirements (e.g., GDPR, HIPAA). TabPFN is also available via private cloud marketplaces.

TabPFN-2.5 outperforms all other methods and complex ensembles tuned for hours on datasets up to 50K samples and 2K features. It also elimates much of the preprocessing, feature engineering, and hyperparameter tuning required in traditional pipelines. This means faster iteration cycles and reduced ML engineering overhead. Read the TabPFN-2.5 model report for more details.

TabPFN is optimized for small- to medium-sized datasets. TabPFN-2.5 is the state-of-the-art for datasets up to 50K samples and 2K features. For very large datasets, it can be combined with sampling or hybrid approaches to ensure both performance and efficiency. Get in touch with us for TabPFN Enterprise if your dataset is larger at sales@priorlabs.ai.

Your data needs minimal preparation, as TabPFN can handle raw tabular data while automatically managing missing values, categorical features, text features and numerical features. It is also robust to outliers and uninformative features and can deal with covariates for time-series. That said, pre-processing will boost performance as we see a tuned TabPFN outperforms TabPFN without any tuning.

TabPFN is highly robust to noisy data, including outliers and uninformative features, because its pre-training on millions of synthetic datasets taught it to handle these real-world challenges.

While the core model is robust to general distribution shifts, the specialized time-series extension specifically handles complex time-series patterns with special features on seasonality.