TabPFN-3.5 released, our best model yet. Learn more ->

One Model, Infinite Predictions

Pre-trained tabular foundation models for making predictions on structured data.
Get best-in-class results in seconds - no tuning, no ML pipelines

8K GITHUB STARS

PUBLISHED IN NATURE

TabPFN–3.5

New

Our frontier model for toughest prediction tasks on any data - grouped, temporal, high-cardinality and more. Supercharged by Thinking mode.

Output

Python

>>> tabpfn.predict(X_test)
array(['No', 'Yes', 'No', 'Yes', 'No'])
>>> tabpfn.predict_proba(X_test)
array([[0.92, 0.08],
       [0.15, 0.85],
       [0.78, 0.22],
       [0.03, 0.97],
       [0.88, 0.12]])

TabPFN-3.5 by the numbers

Average accuracy across 51 OpenML datasets

Default

Tuned (4 hours)

Tuned + Ensembled (4 hours)

99% winrate over classic ML with Thinking

On TabArena, leading benchmark for structured data tasks

#1 for real-world prediction tasks

On BeyondArena, a benchmark for production data science

0.2s for predictions on 1M samples

Fast inference for latency-critical predictive workflows

Read Model report

Read Model report

Built for every type of data

Free text, read as a feature

cracked bracket, water ingress at the seal, second visit customer reports intermittent fault under load, no code logged returned item, outer packaging damaged in, cracked bracket, water ingress at the seal, second visit customer reports intermittent fault under load,

1000s distinct IDs, no encoding

SKU-448210 SKU-091774 SKU-233081 SKU-560412 SKU-118935 SKU-702884 SKU-315097 SKU-880123 SKU-046621 SKU-579340SKU-880123 SKU-046621 SKU-579340 SKU-702884 SKU-315097 SKU-880123 SKU-046621 SKU-579340 SKU-880123 SKU-046621

100s of measurements per row

0.995 0.135 0.485 0.412 1.131 1.047 1.349 0.222 0.691 0.142 0.406 0.807 0.137 0.378 1.010 0.863 0.409 0.925 1.233 0.109 1.228 1.077 0.576 0.318 1.440 0.571 0.230 0.235 1.286 0.945 0.995 0.135 0.485 0.412 1.131 1.047 1.349 0.222 0.691 0.142 0.406 0.807 0.137 0.378 1.010 0.863 0.409 0.925 1.233 0.109 1.228 1.077 0.576 0.318 1.440 0.571 0.230 0.235 1.286 0.945

churn_risk

0.94

One forward pass. No feature engineering.

Trusted across industries

This is some text inside of a div block.

40% Improvement in anomaly detection to advance predictive maintenance.

Isabel Fernando, Innovation Manager, Hitachi Rail

Read case study

This is some text inside of a div block.

TabPFN has the potential to fundamentally transform how banks approach prediction tasks.

Maksims Volkovs, Senior Vice President and Chief AI Scientist, Layer 6 AI at TD

Read case study

This is some text inside of a div block.

This collaboration enables us to derive actionable clinical insights from complex proteomic datasets.

Simon Meier, CEO, Oxford Cancer Analytics

Read case study

This is some text inside of a div block.

90% accuracy at cancer patients identification using peripheral blood

Michael Goldberg, PhD, VP R&D at BostonGene

Read case study

This is some text inside of a div block.

22 of 24 outcomes correctly predicted

Surgeon Commander Tim Scott, Consultant Anaesthetist, UHNM NHS Trust

Read case study

This is some text inside of a div block.

TabPFN presents an exciting opportunity to adopt tabular foundation models and unlock their potential to serve more customers with greater precision.

Dr. Dietrich Eherler, Head of Statistical Analysis and Development, Creditplus Bank AG

Read case study

This is some text inside of a div block.

TabPFN performs surprisingly well with minimal training data. This capability is particularly valuable for fraud detection, where labeled examples are limited.

Read case study

This is some text inside of a div block.

Old models were worse than TabPFN in MAPE. Now, our spend forecasting is more accurate and less manual.

Hendrik Dohme, Data Scientist, Exito

Read case study

This is some text inside of a div block.

Production-grade predictions - typically in seconds.

Dael Williamson, EMEA CTO, Databricks

Read case study

For Businesses

Finance

Healthcare

Industrials

Energy

Tech

Algorithmic traders forecasting directional price movements of assets.

Predicting automotive finance fraud cases and reducing bad losses.

Credit risk modelling for mortgage approvals and SME banking.

How TabPFN is Transforming Financial Risk Management

Read our case study

Open source models used at

Cancer risk prediction in hospitalized patients based on biomarkers.

Classifying genetic variants leading to personalized medicine.

Alzheimers disease prediction for hospitalized patients.

TabPFN helps clinicians predict intubation - before it’s too late

Read our case study

Open source models used at

Gas pipelines steel grade classifications for compliance.

Predictive maintenance for machinery using inspection data.

Quality control in manufacturing through automated defect detection.

Read our case study

Open source models used at

Power and electricity traders using time-series forecasting for arbitrage trading.

Energy providers classifying incoming meter data for customer billing.

Monitoring grid performance to identify inefficiencies and anomalies.

Read our case study

Open source models used at

Predicting customers likely to churn in the next 180 days.

Accurately predicting media spending for marketing campaigns.

Generating highly accurate synthetic survey data based on existing samples.

How Exito Transformed Media Spend Forecasting with TabPFN

Read our case study

Open source models used at

Secure & Compliant

Prior Labs is commited to maintaining the highest security and data privacy standards.

Get started

Predictions, where you need them

TabPFN fits into your stack - open source, API, private cloud, or agents. Same frontier model, on your terms.

API

Get best results with API-exclusive Thinking mode.

VPC

Deploy TabPFN-3.5 inside your own cloud environment.

Agents

Supercharge your agents with structured data intelligence.

Start building with TabPFN today