#1 Tabular Foundation Model

Get better predictions now

TabPFN-3.5 is the most accurate model for your tabular data.
Try it on your own table today.

PUBLISHED IN NATURE
#1 ACROSS LEADING BENCHMARKS
5M DOWNLOADS
OPEN-SOURCE

Trusted by industry leaders solving real prediction problems.

Think your current model is good?
Put TabPFN to the test.

Run TabPFN-3.5 on your own data and compare the results with the model you're already using.

5M+ downloads

And counting

#1 across benchmarks

Independent evaluations

99% win rate

Against classical ML

Built for
tables

Not another LLM

TabPFN ranks at the top for a reason

Here’s what’s behind the #1 ranking across benchmarks.

Drop in your messy, real-world dataset

Use numerical, categorical, missing, text-rich, high-cardinality, and wide data.

Get predictions in a single forward pass

Go from your data to predictions without task-specific training or extensive tuning.

Use the same model for all your predictions

Use TabPFN-3.5 for classification and regression across use cases and industries.

Deploy wherever the data needs to stay

Run TabPFN 3.5 locally, in your VPC, or through the API. Choose the setup that works for you best.

We could brag, but they do it better

“Zero-shot inference surpassing my carefully tuned models, consistently taking the top rank in my backtests. TabPFN proved we could move to an inference-only foundation model.”

“TabPFN outperforms gradient boosting on real-life insurance datasets without any tuning. And you get prediction confidence intervals, which are super useful for risk applications.”

“Production-grade predictions in a single forward pass, typically measured in seconds.”

“I spent more than 2 weeks developing a forecast model. TabPFN was right off the bat at my level of accuracy. To say I was impressed is a gross understatement.”

Published in Nature

Put frontier research to work on your own data.

Complex data in. Predictions out in seconds.

No manual encoding. No feature engineering. One forward pass.

Free text, read as a feature

Use unstructured text directly alongside your tabular data.

Thousands of distinct IDs, no encoding

Handle high-cardinality features without manually encoding thousands of categories.

Hundreds of measurements per row

Work with high-dimensional data without manually selecting or engineering features.

Get Started Today.

State-of-the-art predictions, straight out of the box. No lengthy training or hyperparameter tuning required.

Ready when you are

See for yourself.

Take the data you already have. Put TabPFN-3.5 against the model you already trust.