Models · Updated 7 Oct, 01:27 am IST

Google DeepMind releases EmbeddingGemma 2, a multimodal on-device embedding model

Why it matters for readers: It lets you search across photos, voice notes and video using the same model, keeping data on your device.

EmbeddingGemma 2 is a 740 million parameter model designed for on-device multimodal embeddings.1 2
The model natively maps combinations of text, images, audio, video and code into a unified embedding space.1 2
EmbeddingGemma 2 is built on the Gemma 4 architecture and uses the same underlying technology as Gemini Embedding models.1 2
Google released the model under a commercially permissive Apache 2.0 license and positioned it for privacy‑friendly, on-device use cases like search and RAG.1 2

Get a brief like this every morning

Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.

6 Oct, 09:33 pm · Hacker News (100+ points)
EmbeddingGemma 2: An open, lightweight multimodal embedding model
7 Oct, 01:27 am · Google DeepMind
EmbeddingGemma 2: an open, lightweight multimodal embedding model