Google claims EmbeddingGemma 2 outperforms rival embedding models twice its size

Google has released EmbeddingGemma 2, an open-source multimodal model designed to convert text, images, video, audio, and code into vectors. With 740 million parameters and a small memory footprint, the model is optimized for on-device deployment. By mapping diverse data types into a unified space, the model allows developers to build efficient search and retrieval systems that perform on par with significantly larger competitors. Its release under the Apache 2.0 license lowers the technical barriers for running complex multimodal tasks locally on hardware with limited resources.