Introducing Binary Embeddings for Titan Text Embeddings model in Amazon Bedrock

Amazon Titan Text Embeddings V2 now supports Binary Embeddings. With Binary Embeddings, customers can reduce the storage cost for their Retrieval Augmented Generation (RAG) applications while maintaining similar accuracy of regular embeddings.

Amazon Titan Text Embeddings model generates semantic representations of documents, paragraphs, and sentences, as 1,024 (default), 512, or 256 dimensional vector. With Binary Embeddings, Titan Text Embeddings V2 will represent data as binary vectors with each dimension encoded as a single binary digit (0 or 1). This binary representation converts high-dimensional data into a more efficient format for storage in Amazon OpenSearch Serverless in Bedrock Knowledge Bases for cost-effective RAG applications.

Binary Embeddings is supported in Titan Text Embeddings V2, Amazon OpenSearch Serverless and Amazon Bedrock Knowledge Bases in all regions where Amazon Titan Text Embeddings V2 is supported. To learn more, visit the documentation for Binary Embeddings.