BGE Reranker V2 Large
Open-weight cross-encoder for multilingual text reranking
BGE-Reranker-V2-M3 from BAAI is a state-of-the-art cross-encoder reranker supporting 100+ languages. Takes a query and a list of candidate texts and returns a relevance score for each. Top performer on MS MARCO and BEIR benchmarks.
When to use:
- Second-stage reranking after embedding-based retrieval
- Semantic search quality improvement
- When data privacy requires a self-hosted open-weight model
Input:
- Query: Text query to rank candidates against
- Candidate Items: List of text candidates to score
Output:
- Scores: Relevance score per candidate
- Ranking: Candidate indices sorted by descending relevance
Inference Settings
No inference-time settings. Scores are computed deterministically.