Cross-encoder scoring
According to the model card, the reranker consumes the question and document together and directly outputs similarity rather than embeddings.
Open Source Model Profile · BAAI
bge-reranker-base
bge-reranker-base is a 278M-parameter cross-encoder reranker from BAAI. According to the model card, it scores question-document pairs directly and re-ranks top-k documents returned by embedding models.
- Publisher
- BAAI
- Task
- text-classification
- Model type
- xlm-roberta
- License
- mit
- Library
- sentence-transformers
- Publication status
- Accepted · not indexed
Model overview
bge-reranker-base is published by BAAI as a text-classification reranker in the FlagEmbedding family. Captured Safetensors metadata reports 278,044,931 parameters, about 278M. According to the model card, the cross-encoder takes a question and document as joint input and outputs similarity directly instead of embeddings, and the publisher recommends it for re-ranking top-k retrieval results.
Recorded capabilities
Top-k reranking role
According to the model card, it is designed to re-rank the top-k documents returned by an embedding model for the final result set.
Publisher-reported rerank scores
According to the model card's table, this model averages 65.42 across T2Reranking, MMarcoReranking, and CMedQA tasks.
Commercial-friendly MIT terms
According to the model card, FlagEmbedding is under the MIT License and released models may be used commercially free of charge.
Use cases in the source record
- Second-stage reranking where a bge embedding model retrieves the top 100 documents and this reranker distills them to a final top 3.
- Accuracy recovery flows where the card recommends the cross-encoder when a fine-tuned embedding model alone is not accurate enough.
- ONNX and infinity_emb deployment using the card's documented ORTModelForSequenceClassification and AsyncEmbeddingEngine patterns.
Limitations and unknowns
- No independently measured evaluation results were extracted; table scores are publisher-reported card claims.
- No context-window value was extracted from this record.
- The card's usage examples partly describe sibling bge embedding models rather than this reranker checkpoint.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: BAAI/bge-reranker-base
Captured: Unknown. Processed: 2026-09-07T19:34:30.320105+00:00.
We have updated the new reranker , supporting larger lengths, more languages, and achieving better performance. FlagEmbedding Model List | FAQ | Usage | Evaluation | Train | Citation | License More details please refer to our Github: FlagEmbedding . English | 中文 FlagEmbedding focuses on retrieval-augmented LLMs, consisting of the following projects currently: Long-Context LLM : Activation Beacon Fine-tuning of LM : LM-Cocktail Embedding Model : Visualized-BGE , BGE-M3 , LLM Embedder , BGE Embedding Reranker Model : llm rerankers , BGE Reranker Benchmark : C-MTEB News 3/18/2024: Release new rerankers , built upon powerful M3 and LLM…
F001F002F003F004F005F006F007F010F016F025F026F028F029F030F032F033F035F036F037F040F041