Cohere launched Embed 4, a multimodal embedding model that beefs up enterprise search and retrieval for AI apps.
According to Cohere, Embed 4 can quickly search unstructured data including PDF reports, presentation slide and other documents with text, images, tables and diagrams.
The launch is a fast follow-up to Command A, a model designed to minimize compute resources while delivering strong performance.
Embed 4 also can generate embeddings for documents up to 128K tokens or about 200 pages. The model is also multilingual with more than 100 languages.
If you zoom out a bit, Cohere's moves with Embed 4 highlight a broader product strategy that revolves around enterprise use cases. For instance, Cohere said Embed 4 is "optimized with domain-specific understanding of data from regulated industries such as finance, healthcare, and manufacturing."
Cohere in January launched North, an AI platform designed for streamlining work. Cohere is also developing a version of North for banking.
The company also noted that Embed 4 can be deployed in virtual private clouds or on-premise environments. Cohere's game plan is to address enterprise retrieval augmented generation (RAG), which will be critical to deploying AI agents.
Cohere noted that Embed 4 can search unstructured documents where they reside and represent them in a unified vector. Embed 4 is available on Cohere, Microsoft Azure AI Foundry and Amazon SageMaker for virtual private cloud and on-premises deployments.
With Embed 4, Cohere is addressing multiple areas of the model stack including retrieval as well as prompt augmentation with Rerank and generation with Command A models.
