--- format: "aidr-story-markdown/v1" id: "15e4465d474fd42be5f27f3416162a3a154e704dd679afb9e5d3d4d234f29aa0" canonical_url: "https://aidr.today/15e4465d?lang=en" title: "Google Launches EmbeddingGemma 2 for On-Device Multimodal Search" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-10-13T04:53:28.000Z" category: "Data" topics: ["google","inference","data-engineering","llm"] source_urls: ["https://huggingnews.com/ai/google-launches-embeddinggemma-2-for-on-device-multimodal-search-717a2440","https://x.com/MiaAI_lab/status/2109771831569727655","https://x.com/TheTuringPost/status/2109817905177378908","https://marketbrief.now/ai/google-launches-embeddinggemma-2-for-on-device-multimodal-search-717a2440"] summary: "Google has released EmbeddingGemma 2, a 740M parameter AI model that enables natural language or sketch-based searches through video and audio files on local hardware. The system uses Matryoshka embeddings to reduce vector storage to 1/3 while maintaining approximately 95% of retrieval quality. The model supports cross-modal retrieval and is currently the top trending model on Hugging Face." --- # Google Launches EmbeddingGemma 2 for On\-Device Multimodal Search > [Open the canonical story]() **Published:** 2026-10-13T04:53:28.000Z **Category:** Data **Topics:** google, inference, data\-engineering, llm ## Summary Google has released EmbeddingGemma 2, a 740M parameter AI model that enables natural language or sketch\-based searches through video and audio files on local hardware\. The system uses Matryoshka embeddings to reduce vector storage to 1/3 while maintaining approximately 95% of retrieval quality\. The model supports cross\-modal retrieval and is currently the top trending model on Hugging Face\. ## Sources - [Story source]() - [Story source]() - [Supporting source]() - [Story source]()