EmbeddingGemma 2: an open, lightweight multimodal embedding model

Google has released EmbeddingGemma, a lightweight, multimodal model designed to represent text and images as numerical vectors for search and retrieval tasks. Built on the Gemma 2 architecture, the model is available for open research and commercial use. This release provides developers with a smaller, more efficient alternative for applications requiring semantic matching across different media types, such as improved information retrieval systems or visual data organization.