SAP aims to displace the more generic models of vast languages by releasing its own basic “tabular” model, which the company says will reduce training requirements for enterprises.
The model called SAP RPT-1 is a pre-trained model with ready-to-use business and enterprise knowledge. SAP calls this the relational base model, which means it can make predictions based on relational databases even without tuning or additional training.
Walter Sun, global head of AI at SAP, told VentureBeat that the value of the up-to-date model lies in its ability to perform various enterprise tasks, such as predictive analytics, right out of the box.
“Everyone knows language models, and there are some good ones already,” Sun said. “But we trained the model on business transaction data, mainly Excel spreadsheets, so we have a model that can perform predictive analytics where the value is that it is ready to use, which means you don’t need to know the specifics of the business to perform tasks analogous to a language model.”
Sun said RPT-1 could immediately build an enterprise business model based on knowledge gained from decades of data that SAP has collected. Organizations can plug the model directly into the application without even additional tuning.
RPT-1, SAP’s first major AI model family, will be generally available in the “fourth quarter of 2025.” and will be implemented via SAP’s AI Foundation platform. While the RPT-1 is currently available, the company stated that additional models will be available soon, including a state-of-the-art open source model.
SAP will also provide a no-code play environment where you can experiment with the model.
Tabular models and LLM
Tabular or relational AI models learn from spreadsheets, unlike LLMs which learn from text and code. RPT-1 not only understands numbers and relationships between different cells, but is also able to provide more structured and precise answers.
When enterprises decide to employ RPT-1, they can give the model more direction with a bit of context engineering because the model is semantically aware and learns from how it is used.
SAP researchers first proposed the idea that tabular models could both demonstrate semantic awareness and learn from content via an article published in June. ConTextTab was proposed to introduce context-aware pre-training. It uses semantic signals, such as table headings or column types, to guide model training, allowing the model to build a relational structure to the data. It is this architecture that makes the model best suited for tasks that require precise answers, such as financial or enterprise applications.
RPT models rely on the work of ConTextTab, which allows them to learn structured business data, for example from a SAP knowledge graph, and then be able to add more context through consumption.
SAP researchers tested ConTextTab in benchmarks, saying it was “competitive” with similar models such as TabPFN and TabIFL.
More and more industry models
Many enterprises prefer to tune a generic LLM such as GPT-5 or Claude to essentially train the model to only answer questions relevant to their business. However, a shift towards industry-specific models began to take root.
Sun said his experience at his previous company building a very narrow, highly customized AI model for sentiment analysis greatly influenced what sets RPT-1 apart.
“It was a very customized model, a narrow model that requires detailed feedback on specific products, but it wasn’t scalable,” Sun said. “When LLMs came out, this one model measures sentiment. But there are use cases that we can apply that LLM can’t do.”
He said these employ cases include predictions such as determining when a shopper will return to the grocery store, which may involve numerical analysis along with understanding a shopper’s shopping habits. However, some LLMs have begun to integrate with spreadsheets, and AI model providers encourage users to upload similar data to teach context. Microsoft up-to-date ones added Copilot capabilitiesincluding the ability to work in Excel. Anthropic he integrated his Claude model with Excel, complementing his Claude to handle finances. Chinese startup Manus also offers data visualization tool that understands spreadsheets, and ChatGPT can create charts from uploaded spreadsheets and other data sources.
However, SAP noted that it is more than just reading a spreadsheet; RPT-1 should stand out from the competition because it requires less additional information about the company to provide an answer.