Fundamental has released NEXUS, a large tabular model designed for structured data analysis, now available on Amazon SageMaker JumpStart. This foundation model is pre-trained on over 10 billion rows of tabular data, allowing it to perform predictions on enterprise datasets without the manual feature engineering typically required by traditional machine learning. By utilizing a deterministic architecture, NEXUS produces consistent, reproducible results, overcoming the non-deterministic nature of standard large language models.

The model offers several specialized features for structured data, including permutation invariance, which ensures performance remains stable regardless of column order, and cross-schema reasoning to connect data across disparate tables. NEXUS also includes autonomous data cleaning capabilities and supports massive datasets without the need for truncation. The system is designed to handle multi-dimensional relationships across tables, such as predicting customer churn based on transaction frequency, support history, and economic indicators.

To deploy NEXUS on Amazon SageMaker AI, users subscribe to the model package via AWS Marketplace and deploy it onto an ml.p5en.48xlarge instance featuring 8 NVIDIA H200 GPUs. The deployment utilizes the Fundamental Python SDK, which provides a scikit-learn compatible interface for training and inference. The workflow involves uploading data to Amazon S3, where the model performs automated cleanup and feature extraction using standard Python commands like clf.fit and clf.predict.

NEXUS is available in a base version or industry-specific variants tailored for finance, healthcare, manufacturing, and retail. Financial applications include fraud detection and credit risk modeling, while healthcare use cases span clinical trial matching and patient risk stratification. In manufacturing, the model supports predictive maintenance and demand forecasting. The integration with AWS ensures that all data remains within a single-tenant, network-isolated environment, meeting security requirements such as GDPR, HIPAA, and SOC 2. By leveraging SageMaker AI’s managed infrastructure, enterprises can automate retraining through pipelines and scale predictions to petabyte-level workloads.

Fundamental has released NEXUS, a large tabular model designed for structured data analysis, now available on Amazon SageMaker JumpStart. This foundation model is pre-trained on over 10 billion rows of tabular data, allowing it to perform predictions on enterprise datasets without the manual feature engineering typically required by traditional machine learning. By utilizing a deterministic architecture, NEXUS produces consistent, reproducible results, overcoming the non-deterministic nature of standard large language models.

The model offers several specialized features for structured data, including permutation invariance, which ensures performance remains stable regardless of column order, and cross-schema reasoning to connect data across disparate tables. NEXUS also includes autonomous data cleaning capabilities and supports massive datasets without the need for truncation. The system is designed to handle multi-dimensional relationships across tables, such as predicting customer churn based on transaction frequency, support history, and economic indicators.

To deploy NEXUS on Amazon SageMaker AI, users subscribe to the model package via AWS Marketplace and deploy it onto an ml.p5en.48xlarge instance featuring 8 NVIDIA H200 GPUs. The deployment utilizes the Fundamental Python SDK, which provides a scikit-learn compatible interface for training and inference. The workflow involves uploading data to Amazon S3, where the model performs automated cleanup and feature extraction using standard Python commands like clf.fit and clf.predict.

NEXUS is available in a base version or industry-specific variants tailored for finance, healthcare, manufacturing, and retail. Financial applications include fraud detection and credit risk modeling, while healthcare use cases span clinical trial matching and patient risk stratification. In manufacturing, the model supports predictive maintenance and demand forecasting. The integration with AWS ensures that all data remains within a single-tenant, network-isolated environment, meeting security requirements such as GDPR, HIPAA, and SOC 2. By leveraging SageMaker AI’s managed infrastructure, enterprises can automate retraining through pipelines and scale predictions to petabyte-level workloads.