Google EmbeddingGemma 2

Google EmbeddingGemma 2 is an innovative open-source multimodal embedding model developed by DeepMind. Designed to unify various data types, it maps text, code, images, video, and audio into a shared embedding space, enabling cross-modal search and retrieval from a single query. Its lightweight architecture, with only 740 million parameters, allows it to run efficiently on-device—requiring approximately 567MB of RAM on a Pixel 11 Pro—making it highly accessible for mobile and edge applications. Licensed under Apache 2.0, the model is readily available via Hugging Face and Kaggle, promoting open collaboration and customization. Its improved MTEB Code score from 68.76 to 78.68 highlights its enhanced performance in embedding quality. Whether for research, development, or practical deployment, EmbeddingGemma 2 offers a versatile solution for integrating multimodal data in real-time environments.