Google DeepMind Introduces EmbeddingGemma 2 for Multimodal On-Device Embeddings

Google DeepMind introduced EmbeddingGemma 2, its first natively multimodal open model for on-device embeddings, extending the Gemma embedding line from text to code, images, audio and video in a shared vector space. The release targets local semantic search, multimodal retrieval-augmented generation, classification and clustering, where developers need compact representations of content rather than a conversational model.