Prior Labs builds foundation models for the structured data that runs businesses. Its open TabPFN models make accurate predictions on tables, spreadsheets, and databases in a single forward pass — in seconds, without task-specific training.
Notable milestones from Prior Labs’s journey — funding rounds, launches, and other moments worth knowing.
Eighteen months after founding, SAP agreed to buy the lab — reportedly an ~$1.16B deal — and pledged to invest more than €1B over four years to scale it into a European frontier AI lab for structured data, with Prior Labs continuing to operate independently.
SourceA new "Scaling Mode" removed TabPFN's size ceiling, extending it to enterprise datasets of up to 10 million rows. By then the open model had passed 2.3M downloads and was in production at firms including Hitachi and major global banks.
SourceThe 2.5 release handled 20× more data cells than its predecessor — up to 50,000 rows and 2,000 features — matching the accuracy of heavily tuned AutoGluon and tree-based models while still running in a single forward pass.
SourceBalderton led a €9M ($9.3M) pre-seed, joined by XTX Ventures, SAP co-founder Hans-Werner Hector's Hector Foundation, and Atlantic Labs — plus angels including Hugging Face's Thomas Wolf, Black Forest Labs' Robin Rombach, and Silo AI's Peter Sarlin.
Balderton Capital, XTX VenturesPublished in Nature, TabPFN is a transformer trained on millions of synthetic datasets that predicts on small tables in a single forward pass. In 2.8 seconds it outperformed an ensemble of the strongest baselines tuned for four hours — the first true foundation model for tabular data.
Frank Hutter — the world's most-cited AutoML researcher — left academia with Noah Hollmann and Sauraj Gambhir to commercialise TabPFN, the tabular foundation model born in his lab at the University of Freiburg, with ELLIS/Max Planck director Bernhard Schölkopf among the founding advisors.
