BGE, short for BAAI General Embedding, is a family of open embedding and reranking models from the Beijing Academy of Artificial Intelligence, built mainly for retrieval in RAG systems.
In plain words
BGE started with Chinese and English text embedding models plus big training datasets and a Chinese benchmark. It grew into many models, including rerankers. BGE-M3 is a popular version: it handles more than 100 languages, texts up to 8,192 tokens, and three kinds of retrieval in one model, namely dense vectors, multiple vectors per text, and sparse keyword-like weights.
A quick example
Imagine a company search tool over documents in English, Hindi and Spanish. One BGE-M3 model could embed all of them, so a question in one language finds relevant passages in the others.
Where you'll meet it
- Retrieval and reranking for RAG.
- Multilingual and long-document search.
Related
Sources
- PaperC-Pack: Packed Resources For General Chinese Embeddings, Xiao et al. (arXiv) · read 29 Sept 2026
- DocsFlagEmbedding, BAAI (GitHub) · read 29 Sept 2026
- DocsBAAI/bge-m3, BAAI (Hugging Face) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.