Release dates
— Google launched EmbeddingGemma 2 with Apache 2.0 weights. ↗
EmbeddingGemma 2
An open-weight on-device multimodal embedding model mapping text, code, images, video, and audio into a shared vector space.
Outputs shared 768-dimensional vectors, with Matryoshka Representation Learning truncation to 512, 256, or 128 dimensions.
MODEL NUMBER172
- First public date
- 2026-10-06
- Organization
- Google DeepMind
- Family
- Gemma
01
Specifications
Only published specifications are shown.
- Total parameters
- 740,000,000↗
- Parameter scope
- 740M covers the full multimodal encoder configuration; the text-and-code setup is 270M, with vision and audio encoders loaded as needed.↗
- Context
- 8,192 tokens↗
- Input modalities
- Text · Image · Video · Audio↗
- Output modalities
- Embeddings↗
- Open weights
- Yes↗
- License
- Apache-2.0↗
- Status
- Current
02
Overview
Why it mattered
It extends EmbeddingGemma from text embeddings to unified representations across text, code, images, video, and audio, delivering on-device retrieval in a 740M full multimodal model.
03
Architecture & capabilities
Model capabilities
Multimodal embeddingsOn-device
04
Release history
Google launched EmbeddingGemma 2 with Apache 2.0 weights.
View source05
Related models
Select a card to view the related model.
No related models recorded yet
06
Sources
01
EmbeddingGemma 2: an open, lightweight multimodal embedding modelGoogle DeepMind · Published 2026-10-06 · Checked 2026-10-06
Official · Verified02EmbeddingGemma 2 model cardGoogle AI for Developers · Checked 2026-10-06
Official · Verified03EmbeddingGemma 2: The Developer GuideGoogle Developers · Published 2026-10-06 · Checked 2026-10-06
Official · VerifiedSuggest a correction
Copy this template, add the field, proposed value, and primary source, and share it with the archive maintainer.