BGE Reranker Base

BAAI/bge-reranker-base

published Sep 2023 · updated Jun 2024

BGE Reranker Base is a cross-encoder reranker model that re-ranks top-k retrieved documents to improve relevance ranking for retrieval-augmented generation and search systems.

price

$0.008

/ 1k docs

throughput

532 docs/s

specs

Task	Reranking
Architecture	XLM-RoBERTa
Parameters	278M
License	MIT

about this model

BAAI/bge-reranker-base is a cross-encoder reranking model that directly scores the relevance between a query and a document, outputting a similarity score rather than a vector embedding. It is designed to refine retrieval results by re-ranking the top-k candidates produced by a first-stage embedding model, offering higher accuracy at the cost of increased latency. Built on an xlm-roberta architecture (278 million parameters), the model supports both Chinese and English. Its cross-encoder formulation allows it to capture fine-grained interactions between query and document, making it more discriminative than dense embedding models for ranking tasks. The model is part of the BGE series from BAAI and is licensed under MIT. On the C-MTEB reranking benchmark, the model achieves the following scores:

Dataset	MAP	MRR
CMedQAv1-reranking	81.27	84.14
CMedQAv2-reranking	84.10	86.79
Mmarco-reranking	35.46	34.60
T2Reranking	67.28	77.13

Gigarouter hosts this model as a managed, OpenAI-compatible API, enabling developers to integrate cross-encoder reranking into retrieval pipelines without managing infrastructure. The API accepts query and document pairs and returns relevance scores suitable for re-ranking with minimal integration effort.

best for

·Re-ranking search results in retrieval-augmented generation (RAG) pipelines
·Improving document retrieval precision by re-ranking top-k passages from embedding models
·Bilingual reranking in Chinese and English

FAQ

What is BGE Reranker Base best used for?

It is best for re-ranking documents retrieved by a faster embedding model to improve relevance, commonly in RAG systems and search.

How does BGE Reranker Base compare to embedding models in speed and accuracy?

It is more accurate than embedding models for ranking but less efficient, which is why it is typically used on a small set of top-k results.

What input and output format does the model use?

Input: a pair of query and document text. Output: a relevance score (usually as logits or softmax probability).

How do I call BGE Reranker Base via the gigarouter API?

Use the gigarouter OpenAI-compatible endpoint with your API key, passing the query and documents in the request body.

What license is BGE Reranker Base released under?

It is released under the MIT License.

call it

# rerank documents by relevance; billed per document
curl https://gigarouter.ai/v1/rerank \
  -H "Authorization: Bearer $GR_KEY" \
  -d '{"model":"BAAI/bge-reranker-base","query":"capital of France",
       "documents":["Paris is the capital of France.","Bananas are yellow."]}'

get a key + $25 free →model card ↗all models