Cohere’s Embed v4 multimodal embeddings model now available on Amazon Bedrock

Amazon Bedrock now offers Cohere Embed v4, the latest state-of-the-art multimodal embedding model from Cohere that produces high-quality embeddings for text, images, and complex business documents. This powerful addition to Amazon Bedrock enables enterprises to build AI applications with frontier search and retrieval capabilities.

Traditional embedding models often struggle to understand complex multimodal business materials, such as business presentations and sales and financial reports, requiring extensive data pre-processing pipelines. Embed v4 addresses this challenge by natively processing documents with tables, graphs, diagrams, code snippets, and even handwritten notes. The model handles real-world imperfections such as spelling errors and formatting issues, eliminating the need for time-consuming data cleanup and helping you surface insights from previously difficult-to-search information.

With support for over 100 languages, including Arabic, English, French, Japanese, and Korean, Embed v4 enables global organizations to seamlessly search for information, breaking language barriers. The model is also fine-tuned for industries such as finance, healthcare, and manufacturing, delivering superior performance on specialized documents including financial reports, medical records, and product specifications.

Cohere Embed v4 is available for on-demand inference in US East (N. Virginia), Europe (Ireland), and Asia Pacific (Tokyo), and can be accessed from select public AWS Regions through cross-region inference. Review the Amazon Bedrock Model Support by Regions guide for complete regional availability. To get started, visit the Amazon Bedrock console to request model access. For more information, refer to the Cohere product page and documentation.