Prior Labs brings foundation AI models to structured tabular data

While general-purpose AI models have transformed how text and images are processed, structured numerical data has often remained bound to manual machine learning workflows. Spreadsheets and relational databases represent the core operational record for modern businesses, clinical trials, and financial institutions, yet making accurate predictions on these formats typically requires custom feature engineering and painstaking model tuning. Prior Labs was established to address this precise challenge by developing tabular foundation models capable of reasoning across structured datasets. Founded in Freiburg, Germany, within the ELLIS ecosystem, the company focuses on bringing instant, state-of-the-art predictive capabilities directly to spreadsheets and corporate databases.

The underlying technology behind Prior Labs stems from academic breakthroughs in prior-data fitted networks. Co-founder Frank Hutter, a computer science professor whose foundational paper on the subject was published in the journal Nature, joined forces with computer scientist Noah Hollmann and growth executive Sauraj Gambhir to commercialise the concept. Together, the founding team built TabPFN, a pre-trained tabular foundation model that generates predictions in seconds without requiring traditional hyperparameter tuning or complex machine learning pipelines. The system allows teams to feed in raw data and receive high-precision outputs across binary, multiclass, and regression tasks almost immediately.

Real-world deployments illustrate the flexibility of TabPFN across varied technical domains. In healthcare, researchers at BostonGene used the model to map complex immune system profiles, while Oxford Cancer Analytics collaborated with Prior Labs to improve liquid biopsy analysis and clinical decision making in lung disease. On the industrial side, Hitachi integrated TabPFN into predictive maintenance systems for rail transport to anticipate equipment failures before they occur. In the financial sector, Creditplus Bank implemented the technology to enhance car loan approval decisions, and TD explored enterprise-scale financial forecasting, alongside Exito using it for media spend predictions.

Community adoption expanded rapidly following the release of the open-source software, gathering over four million downloads and eight thousand GitHub stars. To address more complex predictions, Prior Labs introduced TabPFN-3.5, incorporating a dedicated thinking mode to process grouped, temporal, and high-cardinality datasets across enterprise environments. This rapid momentum attracted nine million euros in pre-seed funding led by Balderton Capital, alongside participation from XTX Ventures, Hector Foundation, Atlantic Labs, Galion.exe, and individual investors including Hugging Face co-founder Thomas Wolf and AI researcher Bernhard Schölkopf.

The strategic importance of tabular foundation models reached a milestone when enterprise software provider SAP announced an agreement to acquire Prior Labs, aiming to establish a European frontier AI laboratory around the team. As founders Frank Hutter, Noah Hollmann, and Sauraj Gambhir prepare to share their latest technical updates at Slush 2026, the company continues to advance its open-source and enterprise offerings. By replacing months of manual pipeline building with pre-trained predictive intelligence, Prior Labs has demonstrated that tabular data can benefit from the same foundation model paradigm that transformed natural language processing.

Find out more at priorlabs.ai