TextEmbedding-Gecko

TextEmbedding-Gecko is Google's compact, high-performance vector model: it distills LLM knowledge to deliver MTEB-competitive results with a small 256- or 768-dimension footprint.

Gecko is a versatile text embedding model from Google (available via Vertex AI) engineered for maximum efficiency and performance. It employs an innovative two-step distillation process, leveraging Large Language Models (LLMs) to generate and refine the FRet (Few-shot Prompted Retrieval) training dataset. This method yields a compact model that punches above its weight: the 256-dimension version outperforms many existing 768-dimension models on the Massive Text Embedding Benchmark (MTEB). The larger 768-dimension variant competes effectively with models up to seven times its size, making Gecko a top-tier choice for semantic search, retrieval, and classification tasks.

https://cloud.google.com/vertex-ai/docs/generative-ai/embeddings/get-text-embeddings