TabPFN-2.5

Tabular foundation model for classification and regression tasksTabPFN-2.5 is a tabular foundation model delivering state-of-the-art performance across classification and regression tasks on tabular data. By leveraging in-context learning, the model produces accurate predictions in a single forward pass - no dataset-specific training, no hyperparameter tuning. Simply provide training rows as context. It natively handles missing values, outliers, and categorical features. TabPFN-2.5 is built on a transformer architecture, pre-trained on synthetic tabular datasets. The model supports tabular inputs of up to 50,000 samples and up to 2,000 features. Multiple specialized checkpoints are available, optimized for different data profiles, with predictions returned in seconds.