BGE-Reranker
FreeLightweight multilingual reranker for relevance scoring
FreeFree tier
Inputs: text
About BGE-Reranker
BGE-Reranker is a family of multilingual reranker models developed by BAAI. Unlike embedding models, a reranker takes a query and a passage as input and directly outputs a relevance score (mapped to [0,1] via sigmoid). The model is lightweight, easy to deploy, and supports fast inference. The v2-m3 variant is based on bge-m3 and offers strong multilingual capabilities. It is licensed under Apache 2.0 and available on Hugging Face with over 17 million downloads.
Key Features
Directly outputs relevance score from query and passage input
Multilingual support (Chinese, English, and more)
Lightweight model architecture for fast inference and easy deployment
Open source with Apache 2.0 license
Compatible with sentence-transformers and transformers libraries
Supports inference via Hugging Face Inference Endpoints and Azure deployment
Pros & Cons
Pros
- Lightweight and fast inference
- Strong multilingual performance
- Easy to deploy and integrate
- High download count indicates community trust
- Direct relevance scoring without need for embeddings
Cons
- Requires separate embedding step for initial retrieval
- May need GPU for large-scale batch processing
Best For
Information retrieval and search rankingQuestion answering relevance filteringRetrieval-augmented generation (RAG) pipelinesDocument re-ranking for improving search result quality
FAQ
How does BGE-Reranker differ from an embedding model?
An embedding model converts text into vectors for similarity search, while a reranker takes a query and document as input and directly outputs a similarity score (in [0,1] after sigmoid). Rerankers typically provide more accurate relevance judgments.
Is BGE-Reranker free to use?
Yes, the model is open source under Apache 2.0 license and free to use. There may be costs associated with running inference on cloud platforms.
What languages does the v2-m3 model support?
The v2-m3 model is multilingual, supporting Chinese, English, and many other languages.