Gemini Embedding 2: Our first natively multimodal embedding model

Google launched Gemini Embedding 2, its first fully multimodal embedding model that processes text, images, videos, audio, and documents within a unified embedding space across over 100 languages. Built on the Gemini architecture, the model handles multiple content types simultaneously and supports flexible output dimensions from 768 to 3072, enabling developers to balance performance with storage costs. Early testing shows the model outperforms existing competitors across text, image, and video tasks, establishing new performance standards for multimodal applications.