Databricks Trains DBRX-2 Foundation Model for Enterprise

Databricks has released DBRX-2, a 250B parameter foundation model trained with a focus on enterprise data workloads. The model achieves state-of-the-art performance on SQL generation, data transformation, business analytics interpretation, and structured data reasoning, outperforming general-purpose models on these specific tasks.

DBRX-2 was trained on a curated dataset emphasizing business documents, financial reports, technical documentation, database schemas, and analytical queries. This specialized training gives it superior understanding of business context, enabling it to generate more accurate SQL from natural language and provide more relevant analytical insights.

The model integrates natively with Databricks' Lakehouse platform, allowing it to directly query data warehouses, analyze results, and generate visualizations. Enterprise users can ask questions about their data in natural language and receive answers grounded in their actual datasets, with full audit trails of every data access.

DBRX-2 is available as an open-weight model for on-premises deployment and through the Databricks Model Serving platform. The open release follows Databricks' strategy of making enterprise AI accessible without vendor lock-in, with the model licensed under Apache 2.0.

Databricks CEO Ali Ghodsi noted that general-purpose models often underperform on enterprise data tasks because they were not trained with sufficient exposure to structured data patterns. DBRX-2 addresses this gap, with early enterprise customers reporting 35% improvements in SQL generation accuracy compared to GPT-5.