Amazon Titan Embeddings v2

A high-performance second-generation embedding model optimized for RAG, offering flexible vector dimensions and 4x lower costs than its predecessor.

Amazon Titan Embeddings v2 delivers precision for Retrieval-Augmented Generation (RAG) by mapping text to dense vector representations across 256, 512, or 1024 dimensions. This model supports 100+ languages and features unit vector normalization to improve cosine similarity accuracy for enterprise search. By implementing a smaller 256-bit output, users maintain 97 percent of retrieval accuracy while reducing storage costs by 4x. It is accessible via the Amazon Bedrock API, providing a secure, serverless path for developers to build semantic search and recommendation engines using their own proprietary data.

https://aws.amazon.com/bedrock/titan/