SAP-RPT-1.5 Model Card
SAP-RPT-1.5 Model Card provides a clear, shared understanding of how SAP‑RPT‑1.5 works, what it has been evaluated to do, and how to use it responsibly in production. It summarizes capabilities, performance, limitations, and safety considerations so that data teams, architects, and governance stakeholders can make informed decisions about when and how to rely on the model. SAP‑RPT‑1.5 is SAP’s relational pre-trained transformer, a table‑native foundation model available through SAP Generative AI Hub that uses in‑context learning to deliver accurate, explainable predictive insights from structured business data across banking, finance, insurance, sales, supply chain, and other enterprise scenarios. Download the DocumentModel Card: SAP-RPT-1.5
Operationalize SAP’s relational pretrained transformer model to gain predictive,
explainable insights from enterprise data without building models from scratch.
Card Version: 1.5.0.3
Publication Date: 2026-07-24
Operationalize SAP’s relational pretrained transformer model to gain predictive,
explainable insights from enterprise data without building models from scratch.
Card Version: 1.5.0.3
Publication Date: 2026-07-24
Model card SAP-RPT-1.5 | v1.5.0.3
© 2026 SAP SE or an SAP affiliate company. All rights reserved. See Legal Notice on www.sap.com/legal-notice for use terms, disclaimers, disclosures, or restrictions
related to this material.
Table of Contents
1. General Information ......................................................................................................................... 3
1.1. Model description .................................................................................................................... 3
1.2. How to use and access ........................................................................................................... 3
2. Model Details....................................................................................................................................... 3
2.1. Architecture ................................................................................................................................ 3
2.2. Inputs & outputs........................................................................................................................ 3
2.3. Model variants ........................................................................................................................... 4
SAP-RPT-1.5 ........................................................................................................................................................................................................5
SAP-RPT-1.5-large ..........................................................................................................................................................................................5
SAP-RPT-1.5 with targeted context retrieval ..............................................................................................................................5
3. Performance ........................................................................................................................................ 5
3.1. Performance & prediction quality ..................................................................................... 5
Benchmarking on real-world data ......................................................................................................................................................5
Performance on public benchmarks ................................................................................................................................................6
3.2. Latency ......................................................................................................................................... 7
4. Trustworthy AI..................................................................................................................................... 8
4.1. Explainability ............................................................................................................................. 8
Column-Wise explainability.....................................................................................................................................................................8
Row-Wise explainability .............................................................................................................................................................................8
4.2. Ethics & safety ............................................................................................................................ 8
4.3. Training data............................................................................................................................... 9
5. Further References ........................................................................................................................... 9
Appendix ..................................................................................................................................................... 10
Public benchmarks and baselines ..................................................................................................................................................11
Trademarks .......................................................................................................................................................................................................11
© 2026 SAP SE or an SAP affiliate company. All rights reserved. See Legal Notice on www.sap.com/legal-notice for use terms, disclaimers, disclosures, or restrictions
related to this material.
Table of Contents
1. General Information ......................................................................................................................... 3
1.1. Model description .................................................................................................................... 3
1.2. How to use and access ........................................................................................................... 3
2. Model Details....................................................................................................................................... 3
2.1. Architecture ................................................................................................................................ 3
2.2. Inputs & outputs........................................................................................................................ 3
2.3. Model variants ........................................................................................................................... 4
SAP-RPT-1.5 ........................................................................................................................................................................................................5
SAP-RPT-1.5-large ..........................................................................................................................................................................................5
SAP-RPT-1.5 with targeted context retrieval ..............................................................................................................................5
3. Performance ........................................................................................................................................ 5
3.1. Performance & prediction quality ..................................................................................... 5
Benchmarking on real-world data ......................................................................................................................................................5
Performance on public benchmarks ................................................................................................................................................6
3.2. Latency ......................................................................................................................................... 7
4. Trustworthy AI..................................................................................................................................... 8
4.1. Explainability ............................................................................................................................. 8
Column-Wise explainability.....................................................................................................................................................................8
Row-Wise explainability .............................................................................................................................................................................8
4.2. Ethics & safety ............................................................................................................................ 8
4.3. Training data............................................................................................................................... 9
5. Further References ........................................................................................................................... 9
Appendix ..................................................................................................................................................... 10
Public benchmarks and baselines ..................................................................................................................................................11
Trademarks .......................................................................................................................................................................................................11
Model card SAP-RPT-1.5 | v1.5.0.3
© 2026 SAP SE or an SAP affiliate company. All rights reserved. See Legal Notice on www.sap.com/legal-notice for use terms, disclaimers, disclosures, or restrictions
related to this material.
1. General Information
1.1. Model description
SAP-RPT-1.5 (Relational Pretrained Transformer) is a table-native relational foundation model
built by SAP. It is designed with a deep understanding of the relationships, structures, and
semantics inherent in tabular enterprise data.
The core capability of SAP-RPT-1.5 is tabular in-context learning, where the model makes
predictions based on context rows and query rows that are provided at inference time. SAP-RPT-
1.5 can be applied out of the box across a wide range of enterprise tasks without requiring any
training or fine-tuning steps.
1.2. How to use and access
SAP-RPT-1.5 is available for productive use via generative AI hub on SAP AI Core. If you are an
existing SAP customer, you can follow these steps to access SAP-RPT-1.5 in your SAP BTP
environment. Find more information about SAP-RPT-1.5 here.
To try out the model, you can access the RPT playground free of charge at https://rpt.cloud.sap.
Simply upload your own tabular dataset or use one of the example scenarios provided to get
hands-on with the model.
2. Model Details
2.1. Architecture
SAP-RPT-1.5 employs a transformer-based architecture optimized for tabular data, featuring
semantic embeddings, a table-native 2D attention scheme, and classification/regression heads
for output generation. The model is designed to detect connections and dependencies across
rows and columns in large enterprise tables, with optimized processing for the unique semantics
and data types found in business data. For more information on the architecture, see this peer-
reviewed paper from NeurIPS 2025.
2.2. Inputs & outputs
As an in-context learner, the model accepts labeled examples (context rows) of the predictive
task and query rows in which the values to be predicted are masked by a special token
([PREDICT] in the example below), both simultaneously at inference time. Note that there is no
dedicated training step as the model comes pre-trained to deliver instant predictions.
© 2026 SAP SE or an SAP affiliate company. All rights reserved. See Legal Notice on www.sap.com/legal-notice for use terms, disclaimers, disclosures, or restrictions
related to this material.
1. General Information
1.1. Model description
SAP-RPT-1.5 (Relational Pretrained Transformer) is a table-native relational foundation model
built by SAP. It is designed with a deep understanding of the relationships, structures, and
semantics inherent in tabular enterprise data.
The core capability of SAP-RPT-1.5 is tabular in-context learning, where the model makes
predictions based on context rows and query rows that are provided at inference time. SAP-RPT-
1.5 can be applied out of the box across a wide range of enterprise tasks without requiring any
training or fine-tuning steps.
1.2. How to use and access
SAP-RPT-1.5 is available for productive use via generative AI hub on SAP AI Core. If you are an
existing SAP customer, you can follow these steps to access SAP-RPT-1.5 in your SAP BTP
environment. Find more information about SAP-RPT-1.5 here.
To try out the model, you can access the RPT playground free of charge at https://rpt.cloud.sap.
Simply upload your own tabular dataset or use one of the example scenarios provided to get
hands-on with the model.
2. Model Details
2.1. Architecture
SAP-RPT-1.5 employs a transformer-based architecture optimized for tabular data, featuring
semantic embeddings, a table-native 2D attention scheme, and classification/regression heads
for output generation. The model is designed to detect connections and dependencies across
rows and columns in large enterprise tables, with optimized processing for the unique semantics
and data types found in business data. For more information on the architecture, see this peer-
reviewed paper from NeurIPS 2025.
2.2. Inputs & outputs
As an in-context learner, the model accepts labeled examples (context rows) of the predictive
task and query rows in which the values to be predicted are masked by a special token
([PREDICT] in the example below), both simultaneously at inference time. Note that there is no
dedicated training step as the model comes pre-trained to deliver instant predictions.
Model card SAP-RPT-1.5 | v1.5.0.3
© 2026 SAP SE or an SAP affiliate company. All rights reserved. See Legal Notice on www.sap.com/legal-notice for use terms, disclaimers, disclosures, or restrictions
related to this material.
Figure 1: Input and Output of Tabular In-Context Learning via SAP RPT
SAP-RPT-1.5 supports task types of classification and regression. Schema information including
column names and data types can be provided to allow the model to make best use of the given
data. For details on input and output payloads see SAP-RPT-1.5 on help.sap.com.
2.3. Model variants
SAP-RPT-1.5 is the successor model version to SAP-RPT-1 and can be consumed in three model
variants that mainly differ in their input characteristics:
All model versions allow up to 10 simultaneous prediction columns and 512 query rows (16 in
case of TCR) in each prediction request. Data can be provided as (gzip-compressed) JSON or
1 Planned for general availability in August 2026
2 Generic recommendations on context length are best practice values based on internal tests, balancing prediction quality, model runtime, and costs. Optimal
settings vary across use cases and data sets, so testing with production data is recommended to achieve the best results for a given scenario.
3 The number of target classes is a recommendation for best prediction quality; there is no technical hard limit. Depending on the specific use case, models can
produce high-quality predictions for larger numbers of target classes.
Value SAP-RPT-1.5 SAP-RPT-1.5-large SAP-RPT-1.5 + auto-
context (TCR) 1
Max. accepted context
length
2048 rows 65536 rows unlimited
Max. columns 100 columns 256 columns 100 columns
Max. prediction rows
per API call
512 512 32
Recommended context
length
500–2000 rows 2 4000–8000 rows 1 unlimited
Recommended max. no.
of target classes
(classification only)
~250 3 ~1000 2 unlimited
© 2026 SAP SE or an SAP affiliate company. All rights reserved. See Legal Notice on www.sap.com/legal-notice for use terms, disclaimers, disclosures, or restrictions
related to this material.
Figure 1: Input and Output of Tabular In-Context Learning via SAP RPT
SAP-RPT-1.5 supports task types of classification and regression. Schema information including
column names and data types can be provided to allow the model to make best use of the given
data. For details on input and output payloads see SAP-RPT-1.5 on help.sap.com.
2.3. Model variants
SAP-RPT-1.5 is the successor model version to SAP-RPT-1 and can be consumed in three model
variants that mainly differ in their input characteristics:
All model versions allow up to 10 simultaneous prediction columns and 512 query rows (16 in
case of TCR) in each prediction request. Data can be provided as (gzip-compressed) JSON or
1 Planned for general availability in August 2026
2 Generic recommendations on context length are best practice values based on internal tests, balancing prediction quality, model runtime, and costs. Optimal
settings vary across use cases and data sets, so testing with production data is recommended to achieve the best results for a given scenario.
3 The number of target classes is a recommendation for best prediction quality; there is no technical hard limit. Depending on the specific use case, models can
produce high-quality predictions for larger numbers of target classes.
Value SAP-RPT-1.5 SAP-RPT-1.5-large SAP-RPT-1.5 + auto-
context (TCR) 1
Max. accepted context
length
2048 rows 65536 rows unlimited
Max. columns 100 columns 256 columns 100 columns
Max. prediction rows
per API call
512 512 32
Recommended context
length
500–2000 rows 2 4000–8000 rows 1 unlimited
Recommended max. no.
of target classes
(classification only)
~250 3 ~1000 2 unlimited