Google claims EmbeddingGemma 2 outperforms rival embedding models twice its size

via The Decoder · 4 views · 1d ago

Google just dropped a new AI model that's a true lightweight champion. It's called EmbeddingGemma 2, an open model with 740 million parameters designed to turn just about anything – text, images, video, audio, even code – into vectors. The real kicker? It runs on-device, needs only about 191 MB of RAM, and still outperforms some competing models twice its size. This makes powerful multi-modal AI embeddings more accessible and efficient than ever. It's a major step for AI on your device. Expect smarter, faster on-device AI experiences.

← Back to the feed