Google DeepMind has released EmbeddingGemma 2, an open multimodal embedding model capable of processing text, images, video, and audio into a unified 768-dimensional vector space. This 740M parameter model is designed for consumer hardware, offering low-latency representations for applications like search and RAG. It supports over 100 languages, includes code understanding improvements, and features Matryoshka Representation Learning for reduced storage costs. AI
IMPACT Enables on-device multimodal AI applications with low-latency embeddings and reduced storage costs.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=2 ai=1.0]
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →