E5-large-v2

E5-large-v2 (intfloat/e5-large-v2) is a 335-million-parameter English text embedding model from Microsoft's E5 line, mapping sentences, paragraphs, and documents into 1024-dimensional dense vectors. It became one of the most widely deployed open embedding models for retrieval, semantic search, reranking, and RAG pipelines.

The MIT-licensed weights load through sentence-transformers with a 512-token input limit, and the model remains a default choice in production RAG stacks years after release for its balance of accuracy, speed, and tiny 1.3GB footprint. Date is era-level (2023).