AI Tools & Products
Oct 6, 2026
Google DeepMind launches EmbeddingGemma 2 for multimodal embeddings on devices
Oct 6, 2026
AI Summary
Google DeepMind has introduced EmbeddingGemma 2, an advanced model for on-device multimodal embeddings that integrates text, images, audio, and video. This model, which features 740 million parameters, enhances capabilities for local search and retrieval while prioritizing data privacy and efficiency.
- EmbeddingGemma 2 is designed to unify various data types, including text, images, audio, and video, into a single embedding space for on-device use.
- The model is built on the Gemma 4 architecture and is released under the Apache 2.0 license, featuring 740 million parameters for optimal performance.
- It improves code performance significantly, achieving a score of 78.68 in MTEB Code, and is suitable for tasks like semantic code search and local codebase indexing.
- EmbeddingGemma 2 allows for local generation of embeddings, enhancing data privacy and reducing latency in search and retrieval processes.
- The model can be paired with generative models like Gemma 4 to create efficient on-device retrieval augmented generation (RAG) pipelines.
- Developers can utilize EmbeddingGemma 2 for various applications, including media library searches and video moment retrieval, through tools available in the Google AI Edge Gallery.
- The model supports real-time decision-making capabilities via the MediaPipe Decision Task API, enabling classification and predictive tasks based on multimodal data.
- Google DeepMind collaborated with partners to ensure immediate compatibility and usability of EmbeddingGemma 2 in development environments.
embeddingmultimodalmodelopen sourcelightweight