Leadership and Companies
EmbeddingGemma 2 brings multimodal search to developers’ devices locally
Google launched an open, lightweight model that maps text, images, audio, video and code into one embedding space.
What happened
EmbeddingGemma 2 is a 740-million-parameter multimodal model built on the Gemma 4 architecture and released under the Apache 2.0 licence. It supports on-device search across text, images, audio, video and code, with an 8K-token context window. Its output vectors can be reduced from 768 to 128 dimensions, cutting local storage and memory use by up to six times.
Why it matters
Developers can build privacy-focused search and retrieval systems that process multimodal data locally, reducing dependence on cloud pipelines and enabling offline use.
Source: Google Blog