According to a new press release, Databricks has unveiled DBRX, a groundbreaking large language model (LLM), marking a significant stride in democratizing data and AI. DBRX stands out by surpassing established open-source models on various industry benchmarks, offering organizations worldwide an efficient means to construct and deploy custom LLMs. Unlike relying on closed models, DBRX empowers enterprises to leverage their proprietary data, fostering a new era of customizable AI systems. With a mixture-of-experts architecture, DBRX boasts unparalleled speed and cost-effectiveness in serving, setting a new standard for open-source LLMs.

Developed by Mosaic AI and trained on NVIDIA DGX Cloud, DBRX utilizes a mixture-of-experts architecture, leveraging the MegaBlocks open-source project for enhanced efficiency. The model’s efficiency, coupled with its leading performance, positions DBRX as a catalyst for accelerating the adoption of open-source models over proprietary ones, aligning with the evolving preferences of enterprises seeking transparency and control over their AI endeavors.

Moreover, DBRX is seamlessly integrated into Databricks’ Data Intelligence Platform, offering customers enterprise-grade capabilities for building and deploying generative AI applications. With built-in features for data management, governance, and monitoring, DBRX ensures the safety, accuracy, and governance of AI applications without compromising data control.