Analytics Legends The knowledge platform for SAP Analytics
Academy module

SAP Foundation Models in Practice — SAP-RPT-1 and TabPFN

SAP Foundation Models in Practice — SAP-RPT-1 and TabPFN — Analytics Legends section illustration for the SAP Analytics knowledge base (concepts, studies, Academy)

As of 2026-09-23

Edited by , sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

SAP-RPT and Prior Labs' TabPFN-3.5-Plus predict on business tables without training on customer data: context rows plus [PREDICT] placeholders, one API call. This module covers the model family as of September 2026 (RPT-1 retiring 31 Dec 2026, RPT-1.5, RPT-1.6 with deep context mode, TabPFN-3.5-Plus GA in SAP AI Core), access via the playground, the foundation-models scenario and tabular orchestration, fitting use cases, limits, and a time-based evaluation protocol.

What you will learn

  • Explain tabular in-context learning (context rows, [PREDICT] placeholder, classification vs regression) and why no training on customer data is required
  • Name the SAP-RPT versions and their small/large limits as of September 2026, including the RPT-1 retirement date and RPT-1.6 context_mode
  • Deploy an RPT model through the foundation-models scenario (executable aicore-sap) and call the /predict endpoint or the Python/JS SDK
  • Describe tabular orchestration (Data Destination, Tabular Artifact, Scenario Configuration, /v1/predict) and its model coverage roadmap
  • Run a time-based backtest against a business baseline and decide go/no-go on measured, not claimed, accuracy

Module overview

Most of the value an ERP customer hopes to get from "AI" is not text generation. It is a prediction on a table: will this invoice be paid late, which cost center does this purchase line belong to, how many units will this material sell next month, which supplier is drifting into risk. For a decade the answer was a bespoke machine-learning project — extract, clean, engineer features, train, tune, deploy, monitor, retrain. SAP's tabular foundation models change the economics of that answer. SAP-RPT (Relational Pretrained Transformer) and, since September 2026, Prior Labs' TabPFN-3.5-Plus are pretrained models that make predictions on your rows without training on your data: you send labelled example rows as context and mark the cells to predict, and the model answers in one API call.

Prerequisites

  • Access to an SAP AI Core instance on the extended plan (or the RPT playground for the exercises)
  • Review core concepts first: C165, C107, C026
  • Basic Python or REST client skills; basic classification/regression metrics

Outcomes

  • Explain to a CFO what a tabular foundation model does, what it costs to try, and what it does not replace.
  • Choose between sap-rpt-1.6, sap-rpt-1.6-large, deep context mode and TabPFN-3.5-Plus for a given dataset and latency budget.
  • Produce a working /predict call with context rows, [PREDICT] placeholders and an index column.
  • Write an evaluation protocol with a time-based split, business metrics, a baseline and a confidence-routing rule.

Full module available to members. The full module adds: the decision framework · the end-to-end scenario walkthrough · the KPI scorecard · the anti-patterns · the code blocks · the knowledge check · the diagrams.

Open in the app →

Next in this track