Forget Fine-Tuning: SAP’s RPT-1 Brings Ready-to-Use AI for Business Tasks
SAP aims to replace more common large language models with the introduction of its own fundamental ‘tabular’ model, which the company claims will reduce training requirements for enterprises.
The model, called SAP RPT-1, is a pre-trained model with out-of-the-box business and enterprise knowledge. SAP calls it a Relational Foundation Model, meaning it can make predictions based on relational databases even without refinement or additional training.
Walter Sun, SAP’s Global Head of AI, told VentureBeat in an interview that the value of the new model lies in its ability to perform various business tasks, such as predictive analytics, out-of-the-box.
“Everyone knows language models, and some good ones already exist,” says Sun. “But we trained the model on business transaction data, essentially Excel spreadsheets, and so we have a model that can do predictive analytics where the value is that it’s out of the box, meaning you don’t need specific data from a company to perform tasks analogous to a language model.”
Sun said RPT-1 can essentially build an enterprise business model based on the knowledge gleaned from data from SAP’s decades of information. Organizations can plug the model directly into applications, even without further refinement.
RPT-1, SAP’s first major family of AI models, will be generally available in the fourth quarter of 2025 and deployed through SAP’s AI Foundation. While RPT-1 is currently available, the company stated that additional models will be available soon, including an open-source, state-of-the-art model.
SAP will also release a no-code playground environment to experiment with the model.
Tabular Models vs. LLMs
Tabular or relational AI models learned from spreadsheets, unlike LLMs, which learned from text and code. RPT-1 not only understands numbers and the relationships between different cells, but can also provide more structured and accurate answers.
When companies decide to use RPT-1, they can give more direction to the model with a little context engineering, because the model is semantically aware and learns based on how it is used.
SAP researchers first proposed the idea that tabular models can both demonstrate semantic awareness and learn from content through a paper published in June. It proposed that ConTextTab introduce context-aware pretraining. It uses semantic cues, such as table headers or column types, to guide model training, allowing the model to build a relational structure with the data. It is this architecture that ensures the model works best for tasks with precise answers, such as financial or business use cases.
The RPT models build on the ConTextTab work that allows it to learn structured business data, for example from SAP’s knowledge graph, and then add more context through use.
SAP researchers tested ConTextTab against benchmarks and said it is “competitive” against similar models like TabPFN and TabIFL.
Industry-specific models continue to grow
Many companies prefer to refine general LLMs like GPT-5 or Claude, essentially retraining the model so that it only answers questions relevant to their business. However, there is a shift towards industry-specific models are starting to take root.
Sun said his experience at a previous company building a very limited, highly customized AI model for sentiment analysis influenced much of what makes RPT-1 different.
“It was a very customized model, a limited model that needs specific feedback for specific products, but it was not scalable,” Sun said. “When LLMs came into being, that one model measures sentiment. But there are use cases we can do that LLMs can’t do.”
He said these use cases include predictions, such as determining when a customer will return to a grocery store, which can involve both numerical analysis and insight into the customer’s purchasing behavior. However, some LLMs have started integrating into spreadsheets, and AI model providers are encouraging users to upload similar data to teach them context. Microsoft newly added options for Copilotincluding the ability to work in Excel. Anthropic integrated his Claude model with Excel, in addition to it Claude for the financial services. Chinese startup Manus also offers one data visualization tool that understands spreadsheets, and ChatGPT can create graphs from uploaded spreadsheets and other data sources.
However, SAP noted that it is more than just reading a spreadsheet; RPT-1 should differentiate itself from its competitors because it requires less additional information about a company to provide its answers.
