Prior Labs Releases TabPFN-3.5: A Tabular Foundation Model That Beats the Winning Otto Kaggle Solution With Default Settings

Prior Labs introduced TabPFN-3.5, a tabular foundation model that predicts in a single forward pass without training on the dataset. It scored 0.375 log loss in the Otto 2015 competition, beating the winning result of 0.382, and took first place on 7 benchmarks. Open weights are available for research only; production requires API or commercial license.

• Model grew to 220 million parameters vs 53 million in TabPFN-3
• On Otto 2015, result 0.375 vs winning 0.382 of a 36-model stack
• Run on raw data with default settings took about a minute on RTX PRO 6000
• Supports up to 1 million rows, recommends 6000 features, maximum 20,000