Canadian AI startup Cohere has unveiled Embed V3, a new iteration of its embedding model. The model is designed for semantic search and applications that use large language models (LLMs).

Embed V3 transforms data into numerical representations, referred to as "embeddings.” Its primary features include advanced capabilities in matching documents to queries, increasing the efficiency of retrieval augmented generation, and reducing the operational costs of LLM applications.

It aims to solve some of the challenges of LLMs such as lack of access to updated information and generation of false data. Moreover, the model is compatible with vector compression methods, which can cut down the costs of running vector databases while maintaining high search quality.