Models · Updated 15 Oct, 04:17 pm IST

Google launches EmbeddingGemma 2, a compact multimodal embedding model

Why it matters for readers: It shows how one small model can handle text, code, images, video, and audio for faster, private searches on devices.

EmbeddingGemma 2 is an open-weight, multimodal embedding model released under the Apache 2.0 license.1 2
It projects text, code, images, video frames, and audio into a single 768-dimensional vector space.1 2
The model is modular and scales from 270M parameters (text/code) up to 740M parameters for full multimodal use.1
Designed for on-device use, it can run with about 191MB active RAM for text-only and about 567MB for the full multimodal model on a Google Pixel 11 Pro.2
Without fine-tuning, it supports zero-shot intent routing and can act as a low-latency, on-device decision engine for semantic search and retrieval.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.

Time unknown · Google Developers Blog
EmbeddingGemma 2: The Developer Guide
Time unknown · Google Developers Blog
Bring multimodal semantic search to the edge with EmbeddingGemma 2