SAP Unveils RPT-1, a Tabular Model for Enterprise Business Predictions

SAP is releasing a new enterprise relational foundation model, designed to provide organizations with better insights into their operations. Named RPT-1, the model promises rapid and accurate predictions for non-text-based data, helping businesses anticipate outcomes and make more informed decisions.

This news is part of a bevy of announcements SAP is rolling out as it kicks off its TechEd developer conference. Other AI headlines the company is making include the addition of new Joule agents, updates to Joule Studio, the planned launch of its prompt optimizer, and the fact that it will now locally host large language models (LLMs) on its European cloud infrastructure.

“SAP’s announcements today give the tools they need to deliver at the speed of AI,” Muhammad Alam, a member of the company’s executive board, remarks in a release. “Innovations across SAP’s unique flywheel of applications, data, and AI put developers in the driver’s seat—where they belong.”

Everything About RPT-1

RPT-1 is SAP’s first Relational Pre-Trained Transformer model, which means it’s built to understand and make predictions from relational data, not just text or images. Unlike traditional language foundation models, SAP-RPT-1 is designed for relational and structured business data. It comes pre-trained, eliminating the need for SAP customers to deal with model training tasks that are typically required when using new models.

“We’ve been working on a tabular model, which is basically different from language models in the sense that language models train off of natural language, whereas tabular models trained off of transactions like an Excel spreadsheet, or maybe if you want to look at business supply chain or financial transactions, you have a ledgar,” Dr. Walter Sun, SAP’s global head of AI, explains to me in an interview.

In other words, while language models have many use cases, one area where they struggle is with structured, relational data. This information is often found in business systems, such as CRMs, enterprise resource planning systems, or financial databases—areas where SAP is a significant player. When a tabular model like SAP-RPT-1 is utilized, organizations can run predictive analytics, financial modeling, customer analytics, and scenario simulations to analyze data relationships.

It’s like a fortune-teller, predicting when a customer might make future transactions, identifying potential delivery delays, analyzing hiring trends, and estimating compensation based on market data, calculating the probability of payment defaults, helping forecast new business based on competitors’ historical data, and more.

“We’re creating a model that’s different from, but akin to, language models that can do predictive analytics, extrapolation of data, [and] interpolation of data,” Sun adds.

Sun declined to specify its parameters, though he shares that it is a large model. In truth, there are two versions: SAP-RPT-1-small and SAP-RPT-1-large. Initially, SAP informed me that SAP-RPT-1 was a proprietary model and that there were no plans to open-source it in the near future. However, that was incorrect because the company would announce SAP-RPT-1 OSS, an open-source version on Hugging Face and GitHub.

A company spokesperson interjected to affirm that SAP can create such a tabular model due to its 50-plus-year history, touting that it runs more business-critical applications than anyone else, and therefore recognizes that there’s a data opportunity “just out there for the taking.”

“This business foundation model is trained on data that we have access to, and given that we have products in all lines of businesses, we can actually do predictions in our areas, meaning that SAP has a full business suite,” Sun clarifies. “We actually have data across different lines of businesses, and we also even tie them together with the knowledge graph. So, in theory, we can not only predict more lines of businesses than some companies that don’t have those products, but we can also connect the dots across two different lines.”

Developers can access SAP-RPT-1 exclusively through SAP’s generative AI hub when the model is made generally available later this year—the open-source option is available for download now. In the meantime, the company is releasing a no-code, web-based playground to allow anyone to test it out.

SAP also declares that it plans to release additional SAP-RPT models in the future, though it didn’t provide any specific timelines.

New Joule Agents

Weeks after releasing a new tranche of Joule agents, SAP is introducing five more, but this time they’re designed to automate the management of business processes. Sun describes them as “junior assistants” that are specialized for specific tasks and orchestrate multiple agents. These five bots are all powered by SAP Signavio, its BPM platform to help organizations understand and transform their workflows across the enterprise.

The five new agents are:

• Screen Guide Agent: Helps streamline the user experience by providing live on-screen guidance. The bot accelerates onboarding and can assist users in interpreting complex information quickly and finding value.
• Value Case Creation Agent: Automatically transforms insights gleaned from process mining into structured business cases, identifying any inefficiencies and quantifying the financial impact of recommended changes.
• Dashboard Analyzer Agent: Intelligently interprets complex process mining dashboards and event logs to identify trends and inefficiencies through the contextualization of an organization’s KPIs and simplified data analysis.
• Process Content Recommender Agent: This bot pulls from both SAP’s standard process guidelines and an organization’s custom workflows to answer questions and suggest relevant process resources.
• Workspace Administration Agent: Onboard new users for process modeling and mining faster. It manages enrollment requests as well as licenses, roles, and access level assignments.

All of these are expected to be generally available in the first quarter of 2026.

New Ways to Build Agents in Joule Studio

To make it easier for its customers to create AI agents, SAP has multiple development platforms to choose from. One of those is Joule Studio, similar to Microsoft Copilot Studio and Google’s Vertex AI Agent Builder. Unveiled earlier this year, SAP says it’ll be generally available next month.

But that’s not the only update to Joule Studio. It will soon support system-triggered agents, centralized enterprise-grade agent monitoring, the Agent2Agent interoperability protocol, and Joule Agent extensibility. These features won’t be launching until sometime in the first half of next year.

SAP’s Prompt Optimizer to Launch in Q4 2025

Introduced earlier this year at its Sapphire conference, SAP is acknowledging that its prompt optimizer will be generally available sometime in Q4 2025. However, the release isn’t that surprising since the company initially disclosed it would be “expected later in 2025,” and with two months left in the year, it was only a matter of time.

If you’ve ever wanted to craft the perfect prompt, this tool is perhaps the ideal tool to use. SAP claims that it’s helped reduce days of manual work to minutes. This prompt optimizer tool was built collaboratively with frontier AI lab, Not Diamond, and works across a host of models. For those use cases in which multiple models are used in an app or agent, this prompt optimizer could modify the instructions to generate the best response, all without needing to manually rewrite prompts for ChatGPT, Claude, Mistral, Qwen, etc.

It’s certainly beneficial for those doing prompt engineering.

Safe Harbor Model Hosting in Europe

To satisfy EU data residency requirements, SAP is launching a version of its AI Core foundation in the region. Moreover, it will locally host LLMs on its European cloud infrastructure, a process that began in September. SAP explains this move is part of its initiative to “help organizations adopt new technologies while maintaining control and compliance for their data infrastructure.” The data, AI workflows, and models will be managed and hosted by SAP or its certified partners.

Update as of 11/4/2025 at 8:50 a.m. PT: Post has been edited to reflect that SAP-RPT-1 is not entirely a proprietary model—there are two closed variants and an open-source option.

Featured Image: SAP logo on display at the company's Connect conference in Las Vegas, Nevada on Oct. 6, 2025. Credit: Ken Yeung