Conference Presentation, Product Demonstration
The Foundation Model Revolution for Structured Data | Frank Hutter, Prior Labs | RAISE Summit 2026
- Prior Labs aims to become a world leader in European tabular foundation models by combining deep learning, statistics, and data science, supported by SAP's investment of over one billion euros over the next four years.
- The organization will maintain its independence regarding brand, team, customers, offices, open-weight models, and publications while receiving SAP's direct path to productization and long-term capital.
- TabPFN is predicted to generate predictions in seconds instead of the traditional three to six months, addressing operational inefficiencies where Large Language Models would be approximately a million times too slow due to sequence flattening requirements.
- The technology is expected to lift data science tasks beyond simple prediction to include causal reasoning and interpretability, with a specific capability to generalize well in low-data scenarios by matching the performance of the next best model using only 50% of the data.
- TabPFN is characterized as deterministic with no hallucinations, distinguishing it from LLMs, and is expected to handle distribution changes in new data without requiring retraining or governance updates.
- A synergy between LLMs and TabPFN is anticipated to enable chat-based analytics agents where LLMs handle general statistics while TabPFN executes predictions for higher accuracy.
- For regulated businesses, the distillation of TabPFN into tree-based models is expected to incur quite small performance losses while generating superior models compared to traditional methods.
- The strategy includes moving customers from maintaining hundreds of static models to generating individual, on-the-fly models via agent integration.
- Prior Labs intends to focus on trustworthiness and specialized foundation models to contribute to European sovereign AI, which the outlook deems in urgent need of massive compute investments, open source support, and bold reforms.
- The organization plans to continue publishing open-weight models and research papers to maintain its research-heavy status, having achieved a billion-euro investment commitment within an 18-month timeline from founding.