Natively multimodal
Unifies text, images, video, and audio into a single embedding space, enabling cross-modal search and Retrieval Augmented Generation (RAG) pipelines
EmbeddingGemma
Lightweight, open embedding models optimized for on-device use cases
Search images with a text query. Find a specific moment in a video using a voice memo. EmbeddingGemma 2 enables search, routing, and retrieval systems that can run on everyday devices.
Performance
EmbeddingGemma 2 delivers high-quality embeddings for code, vision, and audio tasks.
Capabilities
Modular-by-design
Get robust performance with the model's 740M parameter form factor, or run only what you need for your task with optional audio and vision encoders
8K token context window
Process long documents, video recordings, or extended audio files up to 5.5 minutes with EmbeddingGemma 2's extended context window
Get started with EmbeddingGemma 2
Hugging Face
Kaggle
