Skip to main content
TabPFN is a pre-trained transformer trained on billions of synthetic datasets to “learn the learning process.” Instead of re-optimizing weights for every new dataset, TabPFN encodes inductive biases, priors, and optimization strategies and applies them to your data via in-context learning. That means one forward pass → high-quality predictions in seconds.

​
How to access TabPFN

API Client

The fastest way to get started with TabPFN. Access our models through the cloud without requiring local GPU resources.

Python Package

Local installation for research and privacy-sensitive use cases with GPU support and a scikit-learn compatible interface.

​
Capabilities

Classification

Solve binary or multi-class classification problems with calibrated probabilities.

Regression

Estimate continuous values with uncertainty-aware outputs and minimal preprocessing.

Forecasting

Model time series (via TabPFN for forecasting) to predict future values and trends.

Anomaly Detection

Detect rare and anomalous samples using TabPFN.

Data Generation

Generate realistic synthetic tabular data with TabPFN.

Fine Tuning

Optimize TabPFN models to your own data with fine-tuning.

​
Why teams choose TabPFN

Accurate predictions in seconds

TabPFN-2.5 reaches tuned-ensemble–level performance with near-instant training.

No re-training required

Skip repeated training loops. Simply update the context and TabPFN performs zero-shot inference.

Familiar interface

Plug into any workflow with the familiar scikit-learn interface or through the Prior Labs API.

Robust in the real world

Handles missing values, outliers, categorical & text features natively.

Minimal preprocessing

Handles missing values, outliers, categorical & text features natively.

Interpretable

Returns calibrated probabilities and integrates SHAP for explainable outcomes.

​
Get Started

Quickstart Guide

Get up and running in minutes with step-by-step instructions