# Google Open-Sources Native Multimodal Embedding Model EmbeddingGemma 2

- Category: Flash (快讯)
- Published: 2026-10-08T11:58:00.000Z
- Source date: 2026-10-08
- HTML: https://www.88z.ai/en/news/flash-20261008-c2878ae41ca73bd7
- Markdown: https://www.88z.ai/en/news/flash-20261008-c2878ae41ca73bd7.md
- Other language: https://www.88z.ai/news/flash-20261008-c2878ae41ca73bd7.md

Google has released EmbeddingGemma 2, built on the Gemma 4 architecture with 740 million total parameters. The model maps text, code, images, video, and audio into a unified embedding space. Its modular design requires only 270 million parameters for text-only tasks, allowing it to run on the Pixel 11 Pro with approximately 191MB of memory after quantization.

## At a glance

- Built on Gemma 4 architecture, supporting unified embedding space mapping for five modalities
- Modular design optimizes resources, occupying only about 191MB memory on Pixel 11 Pro after quantization

## The update

Google has officially launched EmbeddingGemma 2, a native multimodal embedding model. Built on the Gemma 4 architecture, it features a total of 740 million parameters. Its core capability lies in mapping diverse data types, including text, code, images, video, and audio, into a single unified embedding space for efficient cross-modal understanding and processing. EmbeddingGemma 2 utilizes a flexible modular design. For text-only tasks, the model requires only 270 million parameters, significantly reducing computational resource demands. After quantization, the model occupies approximately 191MB of running memory on the Pixel 11 Pro device, demonstrating its efficiency and lightweight advantages for mobile deployment.

## Sources

- [IT之家](https://www.ithome.com/1/010/125.htm) (2026-10-08)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.