Google Releases EmbeddingGemma 2 Open-Source Multimodal Model

Google released EmbeddingGemma 2 on Tuesday, an open-source multimodal embedding model optimized for on-device applications. Built on the Gemma 4 architecture, the model maps text, images, audio, and video into a unified embedding space and is available under an Apache 2.0 license.

The model features 740 million parameters and an 8,000-token context window. This configuration allows local hardware to process up to 5.5 minutes of audio, 29 images, or 58 video frames. To optimize performance, the model utilizes Matryoshka Representation Learning to reduce output vectors from 768 to 128 dimensions, which provides up to a six-fold storage reduction for local vector databases.

EmbeddingGemma 2 is available for download on Hugging Face and Kaggle. It supports various deployment frameworks, including MediaPipe and LiteRT.