SAP adds TabPFN-3.5 Plus to AI Core for business data

SAP has made Prior Labs' TabPFN-3.5 Plus model available in SAP AI Core, bringing it to customers using the company's AI development environment.

The model is designed to make predictions from structured business data stored in tables, a common format for operational and financial records. It can be used for tasks such as cash flow forecasting, payment delay prediction, supplier risk scoring, customer churn assessment and identifying upsell opportunities.

Tabular data remains central to many business decisions because it underpins finance systems, procurement records and customer databases. SAP is positioning tabular foundation models as a distinct class of AI systems for this type of data, rather than the text and image workloads more commonly associated with large language models.

According to SAP, TabPFN-3.5 Plus works on raw business data without the training or tuning often required in conventional machine learning projects. It is intended to handle missing values, mixed data types and inconsistent fields directly, which could reduce the amount of preparation needed before prediction work begins.

The system uses in-context learning to generate predictions from tabular datasets. It can also work with columns containing thousands of distinct values, including product codes and customer identifiers, while handling different data types in the same dataset.

Structured data

The launch follows SAP's acquisition of Prior Labs, completed in July 2026. The deal brought the research team behind the model into SAP, although Prior Labs continues to operate as an independent entity.

SAP had already said it would invest more than EUR 1 billion in building Prior Labs into a frontier AI lab focused on structured data. That focus reflects the company's view that business AI will increasingly depend on systems able to generate predictions from the records held inside enterprise software.

The availability of TabPFN-3.5 Plus in AI Core is part of SAP's wider push into predictive models for business users. Alongside TabPFN-3.5, SAP also pointed to its SAP-RPT model family, which is aimed at generating predictions from labelled business data.

External benchmarking is part of SAP's case for the new model. According to the company, TabPFN-3.5 Plus ranked strongly on TabArena and BeyondArena, which are designed to test prediction quality on real-world tabular datasets.

Those benchmarks matter because structured datasets are often messy and uneven, particularly in large organisations with multiple systems and long histories of data collection. Traditional machine learning methods can require repeated configuration changes, feature engineering and preprocessing before they produce usable results.

Philipp Herzig, Chief Technology Officer, SAP, described those constraints while outlining the thinking behind the model. "Real-world data is rarely perfect. Datasets often contain complex relationships and varying conditions that make traditional machine learning difficult. TabPFN-3.5 is specifically built for these challenges, providing industry-leading accuracy and scalability for tabular data with less manual effort-making it the most effective tabular AI model in the industry to date," said Philipp Herzig, Chief Technology Officer, SAP.

Prediction focus

SAP argues that this approach could shorten the time needed to move from business data to a prediction. The pitch is aimed at customers that want to apply AI to operational forecasting and risk analysis without setting up a full model development cycle for each use case.

Herzig also outlined how SAP sees the role of tabular AI in its broader product strategy. "With TabPFN-3.5 and the SAP-RPT model family, customers get accurate predictions from labelled business data in minutes, with no training required. For us, tabular AI is not a supporting feature of the Autonomous Enterprise, it is the foundation," he said.

The addition of TabPFN-3.5 Plus to SAP AI Core gives SAP another way to embed predictive tools into the software environments where many customers already manage finance, supply chains and customer operations. It also underlines the strategic value the company places on structured data as the basis for business forecasting, scoring and decision support.