---
title: "Google releases EmbeddingGemma 2, an open model that maps text, images, audio and video into one search space"
url: https://anawords.com/story/google-embeddinggemma-2-open-multimodal-embedding-model
published: 2026-10-11T20:23:16+05:30
updated: 2026-10-11T20:23:16+05:30
section: Open Source AI
author: Mayank Sahu
checked: sources checked by AI, approved by editor
version: 1
---

# Google releases EmbeddingGemma 2, an open model that maps text, images, audio and video into one search space

The 740-million-parameter model is free to use under Apache 2.0. Google says its text-only part runs in about 191MB of memory on a phone.

**In short:** Google DeepMind released EmbeddingGemma 2 on 6 October 2026. The open model turns text, code, images, audio and video into vectors in one shared space. It has 740 million parameters, an 8,192-token input and an Apache 2.0 licence.

## What launched

Google DeepMind released EmbeddingGemma 2 on 6 October 2026.[^1] It is an open embedding model: it turns text, code, images, audio and video into lists of numbers, called vectors, in one shared space, so an app can search across all of them at once.[^1][^2]

## Key specs

- **Size:** 740 million parameters for the full model. Text and code alone need 270 million, and developers can load only the parts they need.[^1][^2]
- **Built on:** the Gemma 4 architecture.[^1]
- **Output:** 768-dimensional vectors by default, which can be shortened to 512, 256 or 128 for up to six times less storage.[^1][^2]
- **Input length:** 8,192 tokens, four times the first EmbeddingGemma. Google says that covers about 5.5 minutes of audio, 29 images or 58 video frames.[^1]
- **Licence:** Apache 2.0, which allows commercial use.[^1][^2]

## On a phone

Google says that, when quantised, the text-only model uses about 191MB of active memory on a Pixel 11 Pro, and the full multimodal model about 567MB.[^1]

## Benchmark claims

Google says the model’s MTEB Code score rose from 68.76 to 78.68 compared with the first version, and that it leads multimodal embedding models under 1 billion parameters on code and audio benchmarks.[^1] The developer guide describes the code gain as 14%.[^2] These are Google’s own results.

## Where to get it

Weights are on Hugging Face (google/embeddinggemma-2) and Kaggle.[^1] Google lists support in tools such as sentence-transformers, Transformers, vLLM, Ollama and LM Studio.[^1][^2] It says the first EmbeddingGemma passed 20 million downloads.[^1]

## Who it is for

Developers building search, recommendations or retrieval-augmented generation (RAG) that must run on a device or on modest hardware, especially apps that mix text with images, audio or video.

## Fact ledger

- [Confirmed] Released on 6 October 2026 by Google DeepMind. (source 1,2)
- [Confirmed] 740 million parameters for the full model; 270 million for text and code. (source 1,2)
- [Confirmed] Apache 2.0 licence. (source 1,2)
- [Confirmed] 8,192-token input; 768-dimensional output, shortenable to 128. (source 1,2)
- [Company claim] About 191MB of active memory for the text-only model on a Pixel 11 Pro. (source 1)
- [Company claim] MTEB Code score up from 68.76 to 78.68. (source 1,2)

## Sources

1. [Google: EmbeddingGemma 2 announcement](https://blog.google/innovation-and-ai/technology/developers-tools/embeddinggemma-2/) — official
2. [Google Developers Blog: EmbeddingGemma 2 developer guide](https://developers.googleblog.com/embeddinggemma-2-the-developer-guide/) — official