TLDR Tech / Launches
Google DeepMind releases EmbeddingGemma 2 multimodal model
Google DeepMind has introduced EmbeddingGemma 2, an open-source, compact multimodal embedding model utilizing the Gemma 4 framework. This 740-million-parameter model integrates text, code, images, video, and audio into a unified embedding space, designed for efficient local processing and on-device operation. Released under an Apache 2.0 license, it offers an 8K token context window and incorporates Matryoshka Representation Learning for enhanced storage efficiency.
Read the full story