Gemini Embedding

A multimodal embedding API model introduced by Google that uses a single backbone network to process all modalities. It applies contrastive training techniques to map multiple modalities into a unified vector space, enabling cross-modal retrieval and representation.