gemini-embedding-001

Gemini-embedding-001 is Google's premier, multilingual text embedding model, delivering state-of-the-art vector representations for semantic search and RAG systems.

This is the flagship text embedding model for the Gemini API and Vertex AI, consistently holding a top spot on the Massive Text Embedding Benchmark (MTEB) Multilingual leaderboard. The model supports over 100 languages and accepts inputs up to 2,048 tokens. It generates dense vector representations (default 3072 dimensions) and utilizes Matryoshka Representation Learning (MRL), allowing developers to scale the output to smaller dimensions (e.g., 1536 or 768) for optimized storage and efficiency. Key applications include high-precision semantic search, text classification, clustering, and enhancing Retrieval Augmented Generation (RAG) systems.

https://ai.google.dev/gemini-api/docs/embeddings