TabPFN-3-Plus-Thinking

TabPFN-3-Plus (+Thinking) is the full-capability SKU of TabPFN-3-Plus by Prior Labs. The deployed endpoint serves two modes on the same resource, selected per request. Base TabPFN-3-Plus is a single calibrated forward pass that tops every prior tabular foundation model on the public TabArena benchmark and scales to 1M training rows at 200 features. Thinking mode applies additional inference-time computation on top of TabPFN-3-Plus to push prediction quality further; mode is selected per request via the optional thinking_effort parameter (omit it to fall back to the base forward pass). On TabArena, Thinking mode beats every non-TabPFN model by over 200 Elo overall and by 420 Elo on the largest data subset (10k-100k rows), outperforming the strongest tuned and ensembled baselines at a fraction of their runtime, without using LLMs, real data, internet search, or any external model besides TabPFN. Both modes inherit every base TabPFN-3-Plus capability: state-of-the-art accuracy on classification and regression, native many-class classification, native handling of mixed feature types (numerical, categorical, and free-text string-valued columns), missing values, and outliers. Native text-feature support means string-valued columns are accepted directly, without requiring upstream featurization, and are encoded jointly with the numeric and categorical columns inside the model. Released under the TABPFN-3.0 License v1.0; commercial licenses via Prior Labs.