EmbeddingGemma 2: open multimodal embeddings model for on-device use
Melvin Vivas · X post · 2026-10-07 · Open on X
Topics: Embeddings & Vector Databases, Industry Trends & Job Market · Level: intermediate
Summary
The creator shares Google's announcement of EmbeddingGemma 2, an open embeddings model built to run on-device, and says Gemma releases excite him more than Gemini ones. The model embeds code, images, video, audio and text, comes in sizes from 270M to 740M parameters, supports Matryoshka embeddings to save storage, and uses the Apache 2 license.
Key points
- EmbeddingGemma 2 is an open embeddings model built for on-device use.
- It embeds five kinds of input: code, images, video, audio and text.
- The model is modular, with sizes from 270M to 740M parameters.
- It supports Matryoshka embeddings, so you can shorten the vectors to save storage at some cost in quality.
- It's released under the Apache 2 license, so you can use it commercially.
- The creator finds Google's open Gemma models more exciting than Gemini.
Resources mentioned
- EmbeddingGemma 2 · tool · huggingface.co · free
Google's open 740M-parameter multimodal embedding model that puts text, images, video and audio into one shared vector space for on-device use.
Also in: Google AI Edge Foresight: On-Device Meeting Notes with EmbeddingGemma 2 (Melvin Vivas on X · notes), Try EmbeddingGemma 2 multimodal embeddings in Colab (Melvin Vivas on X · notes), A day of work done from a phone with Grok Bot agents (Melvin Vivas on X · notes), EmbeddingGemma 2 Colab Notebook for FAQ Semantic Search (Melvin Vivas on X · notes) and 2 more - Gemma 4 · tool · ai.google.dev · free
Google's family of open-weight models in several sizes, built to run on devices and offline, with multimodal and agentic abilities, and open to fine-tuning.
Also in: Meetily: Open-Source Local AI Meeting Notes App (Melvin Vivas on X · notes), New Gemma Developer Documentation Site (Melvin Vivas on X · notes), A100 80GB GPUs in Google Colab via Google AI Pro plans (Melvin Vivas on X · notes), A day of work done from a phone with Grok Bot agents (Melvin Vivas on X · notes) and 32 more - Gemini · tool · gemini.google.com · free
Google's multimodal AI model family. In this prototype it interprets voice, pointer position and on-screen content, and writes code to act on what the user wants.
Also in: Export Gemini App Web Apps Directly to Google Sites (Melvin Vivas on X · notes), Gemini 4 Argon Demo: Turning a Research PDF into an Animated Storybook (Melvin Vivas on X · notes), Claude in Google Docs, Sheets and Slides: An Alternative to Gemini (Melvin Vivas on X · notes), Multimodal AI Pointer: Combining Voice, Mouse Pointing and Vision with Gemini (Melvin Vivas on X · notes) and 3 more
Try this
- Build an on-device semantic search over mixed media (text, images, audio) with EmbeddingGemma 2, using Matryoshka-truncated embeddings to save storage.
More in Embeddings & Vector Databases
- EmbeddingGemma 2 Colab Notebook for FAQ Semantic Search
- EmbeddingGemma 2 released on Hugging Face
- EmbeddingGemma 2: One On-Device Model for Multimodal Embeddings
- Liquid AI LFM 2.5 Encoder: a CPU-friendly encoder model
- How RAG Finds the Right Document Fast: Graph-Based Vector Search
- Sentence Transformers v6: ColBERT-Style Multi-Vector Models Fully Supported