SAP has announced the availability of TabPFN-3.5, Prior Labs’ new AI model, within SAP AI Core. The company presents this technology as a specialized option for working with tabular data — such as financial information, payments, suppliers, or customers — and generating predictions without needing to train or fine-tune a model for each case beforehand.
TabPFN-3.5 Plus in 20 seconds
- TabPFN-3.5 arrives in SAP AI Core to work directly with structured enterprise data.
- The model uses in-context learning and requires no specific training or prior fine-tuning.
- It can handle missing values, mixed data types, and columns with thousands of distinct values.
- SAP cites the external benchmarks TabArena and BeyondArena to back up its claims about accuracy and scalability.
- Prior Labs became part of SAP in July 2026, after a deal for which SAP had committed more than €1 billion.
The offering targets a type of information present in much of enterprise systems, but that doesn’t fit neatly with the capabilities large language models (LLMs) were designed for. SAP cites examples such as cash-flow forecasting, payment delays, supplier risk assessment, upsell opportunities, and customer churn risk.
While LLMs are designed mainly to work with language and knowledge, tabular foundation models, known as TFMs, specialize in structured data organized into rows and columns.
That difference has practical consequences. In an enterprise system, data can contain incomplete fields, different types of values, or identifiers with thousands of categories. Preparing that material for a conventional machine learning model can require specific transformations and configurations. SAP argues that TabPFN-3.5 is designed to work with that data with less upfront preparation.
A Model Built for Enterprise Data Without Specific Training
TabPFN-3.5 uses in-context learning, a technique that lets the model make predictions from the examples included in the data it receives. According to SAP, this removes the need to train a specific model for each dataset or run successive configuration tests before getting results.
The company also highlights its ability to natively handle missing values and different data types. Another point it raises is how it handles columns that can contain thousands of distinct values, such as product codes or customer identifiers.
This feature is especially relevant in enterprise systems, where tables don’t always have the clean, uniform structure that some traditional machine learning processes require.
SAP doesn’t position TabPFN-3.5 as a general replacement for LLMs. The company places each technology in a different domain: language models are geared toward linguistic information and knowledge, while TabPFN-3.5 focuses on extracting predictions from tabular data.
The goal is for a customer to be able to use information that already exists in their systems to generate predictions without building a training process from scratch for each application.
Possible use cases include calculating supplier-related risks, forecasting payment delays, or estimating potential customer churn. These are tasks where the expected output isn’t necessarily generated text, but a prediction based on multiple structured variables.
SAP Folds Prior Labs Into Its Enterprise AI Strategy
TabPFN-3.5 Plus’s arrival in SAP AI Core also reflects Prior Labs’ ongoing integration into SAP. The German company completed its acquisition of Prior Labs in July 2026.
SAP had previously announced plans to invest more than €1 billion to expand Prior Labs’ capabilities and build a European AI lab focused on structured data. According to the announcement, Prior Labs continues to operate as an independent entity within SAP’s structure.
The deal lets SAP add to its enterprise offering a technology developed specifically for tabular data. The company also mentions its SAP-RPT model family as part of its strategy in this area.
SAP CTO Philipp Herzig argues that real enterprise data tends to contain complex relationships and variable conditions that make traditional machine learning methods difficult to use. His comments are part of the company’s corporate messaging, and SAP uses them to explain how TabPFN-3.5 is positioned within its offering.
To back up its performance claims, SAP points to results from TabArena and BeyondArena, two external benchmarks designed to evaluate prediction models on real data. The company describes TabPFN-3.5 Plus as the most accurate and scalable tabular model currently available, but that claim reflects SAP’s own assessment based on those benchmarks and shouldn’t be mistaken for a universal feature of the technology.
Access to the model comes through SAP AI Core. For SAP customers, this makes it possible to bring TabPFN-3.5 Plus into the company’s AI services environment rather than separately managing all the infrastructure needed to run the model.
The most relevant question for businesses will be how this type of model fits into their current analysis and prediction processes. The absence of specific training can reduce some of the upfront work needed for certain use cases, though the announcement doesn’t provide general figures on time savings, cost, or accuracy improvements for specific customers.
It’s also worth distinguishing between a model being available and a business decision being fully automated. TabPFN-3.5 can generate predictions on structured data, but how those predictions are used in processes like supplier management, collections, or financial forecasting will depend on each organization and the systems it integrates with.
SAP’s strategy thus points toward AI specialized by data type. Rather than concentrating all tasks in a single generalist model, the company is adding models designed for specific problems and for the data its customers already work with.
Frequently Asked Questions
What is TabPFN-3.5?
It’s a tabular foundation model from Prior Labs designed to make predictions on structured data. SAP has announced its availability as TabPFN-3.5 Plus within SAP AI Core.
Does TabPFN-3.5 need training?
According to SAP, it requires no training or specific fine-tuning to make predictions. It uses in-context learning to work with the data it receives.
What data can TabPFN-3.5 analyze?
SAP says it can work with tabular data that includes missing values, mixed data types, and columns with thousands of distinct values, such as product codes or customer identifiers.
When did SAP acquire Prior Labs?
SAP completed its acquisition of Prior Labs in July 2026. The company had previously announced an investment commitment of more than €1 billion to expand its AI research capabilities for structured data.
