EmbeddingGemma 2 is Google DeepMind's open, lightweight multimodal embedding model. It maps text, code, images, video, and audio into a single shared embedding space, enabling cross-modal search from one query. At 740M parameters it is designed to run on-device, with weights released under the Apache 2.0 license on Hugging Face and Kaggle.
Google EmbeddingGemma 2
An open, lightweight multimodal embedding model
Google EmbeddingGemma 2 Introduction
Key Features
- Multimodal embeddings across text, code, images, video, and audio in one shared space
- Lightweight 740M-parameter model that can run on-device
- Open weights under the Apache 2.0 license, available on Hugging Face and Kaggle
- Optimized for privacy-first use cases
Use Cases
- Cross-modal semantic search across text, images, video, and audio
- On-device applications requiring local, privacy-preserving embeddings
- Retrieval and search features built on open embedding models
Why Startups Use It
Startups can integrate an open, Apache 2.0-licensed multimodal embedding model without heavy infrastructure, enabling privacy-first, on-device search and retrieval features across multiple media types.
Alternative Options
OpenAI Embeddings, Cohere Embed, Sentence Transformers
Frequently Asked Questions
What is EmbeddingGemma 2?
EmbeddingGemma 2 is Google DeepMind's open, lightweight multimodal embedding model that maps text, code, images, video, and audio into a single shared embedding space.
Is EmbeddingGemma 2 open source?
The model weights are released under the Apache 2.0 license and are available on Hugging Face and Kaggle.
Can EmbeddingGemma 2 run on-device?
Yes. At 740M parameters it is designed for on-device use; the full multimodal model requires roughly 567MB of RAM on a Pixel 11 Pro, and text-only use about 191MB.
More About Google EmbeddingGemma 2
PricingPaid
Listed DateOct 13, 2026
Authority Badge
Add our badge to your website to showcase product credibility and listing status.
Featured List