analiza aiqbench

Google has expanded its open-source EmbeddingGemma line to support images, audio, and video within a single unified embedding space, enabling local retrieval of multimedia content on smartphones. The updated model leverages the Gemma 4 architecture with specialized vision and audio encoders while utilizing Matryoshka Representation Learning to significantly reduce memory footprint for vector databases. This advancement allows developers to build privacy-preserving apps that can index and search mixed-media data entirely on-device without relying on cloud infrastructure.