Google launched EmbeddingGemma 2 to unify multimodal on-device search
EmbeddingGemma 2’s modular, lightweight architecture enables developers to build privacy-first, multimodal search and retrieval applications directly on consumer devices without cloud dependency, changing the tradeoff between model capability and deployment constraints.
Frame 1 of 4
Google launches EmbeddingGemma 2, a modular multimodal embedding model
Google DeepMind released EmbeddingGemma 2, a 740 million parameter open model under Apache 2.0 license that natively embeds text, images, audio, and video into a unified embedding space optimized for on-device inference. The model is modular, allowing parameter scaling from 270M for text-only to full multimodal support, and achieves best-in-class benchmark scores among sub-1B models.