TabPFN
Foundation model for tabular data enabling fast local classification and regression.
Run transformer-based tabular classification and regression locally using PyTorch.
TabPFN provides transformer-based foundation models designed specifically for tabular data classification and regression tasks. It suits data scientists and machine learning practitioners who want to perform fast, local inference without traditional hyperparameter tuning.
What it is
The library implements TabPFN classifiers and regressors that download model checkpoints on first use. You can switch between model versions like TabPFN-3.5 and faster variants using simple Python classes.
Who it's for
It suits data scientists working with tabular datasets who have Python environments and PyTorch installed. It does not suit teams lacking access to a GPU for larger datasets or those needing non-Python integrations.
Key features
Classification and Regression
Instantiate TabPFNClassifier or TabPFNRegressor to fit tabular data and predict outcomes.
Multiple Model Versions
Select specific model iterations such as TabPFN-3.5-Fast or older versions using ModelVersion constants.
Extension Ecosystem
Install tabpfn-extensions for SHAP-based interpretability, outlier detection, and synthetic data generation.
Cloud Client Option
Use the tabpfn-client package if you prefer cloud-based inference instead of local hardware execution.
Up and running in 3 steps
Our honest verdict
What's great
- Eliminates traditional hyperparameter tuning workflows for supported tabular tasks
- Provides both classification and regression estimators in a scikit-learn compatible style
- Supports fast local inference on GPUs with automated PyTorch integration
- Includes companion extensions for SHAP interpretations and outlier detection
Where it falls short
- Requires a GPU with adequate VRAM for larger datasets to achieve optimal performance
- Different model versions carry distinct non-commercial weight licenses separate from the code license
- Older model versions and CPU execution impose strict limits on sample sizes
Pricing & license
The code is free under the Apache-2.0 license, but specific model weights are governed by separate non-commercial licenses or prior terms. Paid enterprise options are available for high-throughput production needs.
Open remote AI-training roles
873 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Do I need a GPU to run TabPFN?
A GPU is strongly recommended for optimal performance, though CPUs can handle moderate datasets up to 5,000 samples for newer versions.
Can I use TabPFN for regression tasks?
Yes, TabPFN provides a TabPFNRegressor class alongside its classifier for tabular prediction.
