EmbeddingGemma 2 unifies five modalities in one vector space
Google DeepMind’s open model brings text, code, images, video, and audio into one embedding space. The new AI Edge Foresight app uses it for offline meeting notes and local retrieval on Mac.
AT A GLANCE
Key points
- One 768-dimensional space for text, code, images, video, and audio
- Modular loading from 270 million to 740 million parameters
- Apache 2.0 license and downloadable weights
- Foresight adds offline, on-device meeting notes and retrieval on Mac
What is changing?
Google DeepMind released EmbeddingGemma 2 on October 6 under the Apache 2.0 license. Based on Gemma 4, it maps text, code, images, video, and audio into a shared 768-dimensional embedding space. Its modular architecture starts at 270 million parameters for text and code and reaches 740 million across all modalities, with an 8,192-token context window.
The shared vector space makes it easier to retrieve images, clips, or audio with a text query. Google’s new experimental Mac app, AI Edge Foresight, combines EmbeddingGemma 2 with Gemma 4 in a meeting workflow. It processes microphone and system audio on-device, enriches shorthand notes with conversational context, detects questions, and searches a local personal library for answers while working offline. The app is available through Google Labs.
Explore the original sources
Source date: 6 October 2026.
Published by AI Radar: 6 Oct 2026, 20:03 (Istanbul, UTC+3). Updated: 8 Oct 2026, 17:04.
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…