Google releases 740-million-parameter EmbeddingGemma 2 for on-device use

The model places text, code, images, audio and video in a shared 768-dimensional embedding space and is designed to run on consumer hardware. Released under the Apache 2.0 license, it is an open multimodal model that Google describes as optimized for use cases that prioritize privacy.

• The Next Web reports that the model has no safety tuning.
• Google says performance varies across its 100 supported languages, according to The Next Web.

Covered by 6 publishers within 6 hours of the first report: 4 news reports and 2 from the company.

• The Next Web: Google launches EmbeddingGemma 2, an open multimodal embedding model for devices

Reporting4

UAI Unite.AI DeepMind Debuts EmbeddingGemma 2, Mapping Five Modalities Into One Space · Jonas Reeve, Cognitive AI & AGI, AI Research Agent 12:50 PM

MT MarkTechPost Google DeepMind Releases EmbeddingGemma 2, a 740M Open Multimodal Embedding Model Built on Gemma 4 · Asif Razzaq 2:36 PM

TNW The Next Web Google launches EmbeddingGemma 2, an open multimodal embedding model for devices · Ana Maria Constantin 3:15 PM

SA SiliconANGLE Google expands EmbeddingGemma beyond text to images, audio and video · Duncan Riley 6:18 PM

From the company2

G Google EmbeddingGemma 2: an open, lightweight multimodal embedding model · Sahil Dua 12:00 PM

GDM Google DeepMind EmbeddingGemma 2: an open, lightweight multimodal embedding model 3:57 PM

Discussion

Y Hacker News 229 points, 29 comments

Related

NW Google releases Nano Banana 2.1 to improve image quality and editing October 6

APO Google prepares Gemini to make calls to friends and family October 5

IIC Anthropic adds Claude sidebar to Google's productivity apps October 6

TT Google cuts Gemini model choices for free and AI Plus users October 5

Topics Google