Google DeepMind has released EmbeddingGemma 2, a new open multimodal embedding model capable of processing text, code, images, video, and audio into a single shared embedding space. This 1 billion parameter model supports over 100 languages, features an 8,192 token context window, and is designed for on-device applications like search, retrieval, and RAG systems. EmbeddingGemma 2 outperforms several leading commercial and open-source models on various retrieval benchmarks, including MMTEB v2 and CrossCodeBench. AI
IMPACT Enables on-device multimodal search and RAG applications with a free, open-source model.
RANK_REASON Google DeepMind released a new multimodal embedding model with system card and technical report. [lever_c_demoted from frontier_release: ic=2 ai=1.0]
Read on Mastodon — mastodon.social →
- arXiv
- Crossmodal-3600
- EmbeddingGemma 1
- Gemini Embedding
- Google DeepMind
- Hugging Face
- MMTEB v2
- OpenAI
- text-embedding-004
- text-embedding-3-large
- text-embedding-3-small
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →