The Beijing Academy of Artificial Intelligence (BAAI) develops the BGE embedding model family.

BGE (BAAI General Embedding) is a family of open text embedding models, including bge-m3 and bge-large-en, used for dense retrieval and reranking.

Weights are hosted on Hugging Face under BAAI, with code in the FlagOpen/FlagEmbedding GitHub repository.

First BGE models released in August 2023, with bge-m3 following in 2024.

Built to give the open-source community strong multilingual and long-document embeddings for RAG and search pipelines.

Used via Sentence-Transformers, Hugging Face Transformers, or the FlagEmbedding library.

Most BGE models are MIT licensed, with some variants under Apache 2.0 or Gemma license; fully open weights, no credentials required, self-hosted; maintained by BAAI.