Google Unveils EmbeddingGemma 2 for Multimodal On-Device AI Applications

7 Oct 2026, 12:38 am · 2h ago · 1 min read · OfficeChaiNew

Google has introduced EmbeddingGemma 2, a new open embedding model designed to convert multiple data types into a unified shared representation space. The model supports text, source code, images, video, and audio, allowing developers to generate unified vector embeddings across different modalities directly on edge devices. By standardizing multimodal inputs into a single embedding space, the lightweight architecture facilitates efficient semantic search, content retrieval, classification, and clustering tasks locally on consumer hardware without sending data to external cloud servers. The release expands Google's lightweight open-source Gemma family for on-device AI system development.