Google Launches EmbeddingGemma 2 for Multimodal AI Applications

• •EmbeddingGemma 2 integrates text, images, audio, and video into a single model.
• •It features 740 million parameters and is optimized for on-device performance.
• •The model improves code performance by nearly 10 points compared to its predecessor.

On October 6, 2026, Google DeepMind announced the release of EmbeddingGemma 2, a multimodal embedding model that integrates text, images, audio, and video into a unified embedding space. The model features 740 million parameters and is designed for on-device applications, allowing developers to create privacy-first tools with low latency. EmbeddingGemma 2 improves upon its predecessor with a 9.92-point increase in code performance and supports 100+ languages. It is optimized for consumer hardware, requiring minimal active RAM and offering a flexible architecture with modular encoders. The model also includes Matryoshka Representation Learning for storage efficiency, allowing for dynamic truncation of output vectors. It is released under the Apache 2.0 license, making it accessible for developers. The launch aims to enhance local AI applications without compromising user privacy.