Google releases EmbeddingGemma 2 for on-device multimodal retrieval
Google released EmbeddingGemma 2, a 740M-parameter model licensed under Apache 2.0 that maps text, code, images, video, and audio into a shared 768-dimensional embedding space. Based on Gemma 4, it has a modular design with a 270M-parameter text component and optional 170M-parameter vision and 300M-parameter audio encoders. It supports an 8K-token context window and Matryoshka Representation Learning, allowing embeddings to be truncated to reduce storage. The reports say models are available through Hugging Face and LiteRT; arXiv metadata lists the accompanying paper separately.
The release gives developers an open, compact option for local multimodal search, retrieval and RAG, allowing different content types to be represented in one space. Reports also say shorter Matryoshka embeddings can reduce vector-storage requirements, but involve retrieval-quality trade-offs.
EmbeddingGemma 2: The Developer Guide - Google Developers Blog
EmbeddingGemma 2: The Developer Guide
OCT. 6, 2026
Maarten Grootendorst Member of Technical Staff
Ian Ballantyne Senior Developer Relations Engineer
#EmbeddingGemma 2#Google#Gemma 4#Hugging Face#LiteRT#Kaggle#AI Edge Gallery#AI Edge Foresight#MediaPipe Tasks#MTEB Code#llama.cpp#Unsloth