Google's first natively multimodal embedding model
Gemini Embedding 2 is Google's first natively multimodal embedding model that maps text, images, video, audio and documents into a single embedding space, enabling multimodal retrieval and classification across different types of media and it’s available now in public preview.
The model natively understands interleaved multimodal input and captures semantic intent across 100+ languages, making it ideal for sentiment analysis, data clustering, and classification tasks that span text, images, video, and audio without requiring intermediate transcriptions or format conversions.
Gemini Embedding 2 establishes new performance standards for multimodal embeddings, introducing strong native speech capabilities and outperforming leading models in text, image, and video tasks. It is available in public preview via Gemini API and Vertex AI with integration support for LangChain, LlamaIndex, Haystack, Weaviate, QDrant, and ChromaDB.
Gemini Embedding 2 incorporates Matryoshka Representation Learning (MRL), allowing flexible output dimensions that scale down from the default 3072 to lower dimensions. This enables developers to balance embedding quality against storage and computational costs based on their specific performance requirements.
Gemini Embedding 2 maps text, images, videos, audio, and documents into a single unified embedding space, enabling semantic search and Retrieval-Augmented Generation (RAG) across all media types simultaneously. This eliminates complex multi-pipeline architectures and allows developers to capture nuanced relationships between different modalities in a single request.
The model natively ingests audio without text transcription, processes PDFs up to 6 pages, handles up to 120 seconds of video, and supports up to 6 images per request. This native multimodal processing eliminates preprocessing steps and captures complex relationships that would be lost in format conversions.