Google DeepMind has announced EmbeddingGemma 2, an open, lightweight multimodal embedding model that natively maps combinations of text, code, images, video, and audio into a unified embedding space. The model is built on the Gemma 4 architecture and released under a commercially permissive Apache 2.0 license.
With 740 million parameters, EmbeddingGemma 2 is designed for on-device inference. It follows the original EmbeddingGemma, which surpassed 20 million downloads and was used for on-device search and privacy-first retrieval augmented generation (RAG) pipelines. The new model expands beyond text to support cross-modal search, such as finding a specific video clip from a voice memo or searching hours of audio recordings with a text query.
According to DeepMind, EmbeddingGemma 2 is best-in-class for its size, achieving leading scores among sub-1B multimodal embedders on benchmarks including MTEB Code and MAEB, while matching or outperforming many larger models across text, vision, and audio tasks. It is modular by design, requiring as little as 270M parameters for text-only workloads, with optional vision (170M) and audio (300M) encoders for full multimodal support.
The model uses Matryoshka Representation Learning (MRL) to let developers dynamically truncate output vectors from 768 dimensions down to 512, 256, or 128 dimensions, providing up to 6x storage reduction for local vector databases. On a Google Pixel 11 Pro with quantization, it requires as little as ~191MB active RAM for text-only weights and ~567MB for the full multimodal model. It also features an 8K token context window (4x larger than EmbeddingGemma 1), sufficient for up to 5.5 minutes of audio, 29 images, 58 video frames, or interleaved combinations.
EmbeddingGemma 2 delivers a 9.92-point improvement on code performance in MTEB Code, rising from 68.76 to 78.68, making it suitable for local codebase indexing, semantic code search, and coding agent retrieval. It shares Gemma 4's text tokenizer and audio encoder, allowing both models to run together in a unified pipeline with a lower combined memory footprint.
The model weights are available on Hugging Face and Kaggle, with Gemini Enterprise Agent Platform Model Garden availability coming soon. Developers can deploy via Google AI Edge MediaPipe, LiteRT, transformers.js, or WebGPU, and serve with tools including transformers, sentence-transformers, MLX, vLLM, llama.cpp, SGLang, Ollama, and LMStudio. Fine-tuning guidance is available from Unsloth.