BGE Reranker

bge-reranker-v2-m3 is a reranker model released by BAAI on Hugging Face, built on the bge-m3 base model. Unlike an embedding model, it takes a question and a document as input and directly outputs a similarity (relevance score) instead of an embedding; the score can be converted to a 0 to 1 value with a sigmoid function. It is described as a lightweight model with strong multilingual ability, easy deployment, and fast inference. It can be used with FlagEmbedding, sentence-transformers (CrossEncoder), and Transformers, and is also available through Hugging Face's hf-inference inference provider. The model card license is apache-2.0.

• Directly outputs a relevance score for a query-document pair (convertible to 0 to 1)
• Lightweight reranker with strong multilingual ability (based on bge-m3)
• Easy deployment and fast inference
• Supports the FlagEmbedding, sentence-transformers, and Transformers libraries
• Callable via the Hugging Face hf-inference inference provider
• Guidance on fine-tuning data formats and training methods

Open source + paid cloud-managed plans

Open-source libraries are free, managed vector DB/memory services are paid based on usage

Estimate based on AI knowledge - please verify on the official site

• One of 21 AI tools in the RAG & Memory field.
• Connected to 4 other tools — #7 in its field.