bge-reranker-v2-m3
BAAI's 568M multilingual cross-encoder reranker, the default second stage for open RAG.
About
bge-reranker-v2-m3 is BAAI's multilingual reranker built on BGE-M3: a 568M-parameter XLM-RoBERTa cross-encoder that reads a query and candidate passage together and scores their relevance directly, supporting 100+ languages. It is the standard second stage in retrieve-then-rerank pipelines — pull 20-50 candidates by vector similarity, then let the reranker keep the top few for the LLM's context.
The weights are Apache 2.0 and it runs comfortably on CPU, with GGUF builds and one-line LangChain integration. A reranker adds 50-300ms per query and another model to maintain, so it pays off most when top vector hits are topically right but not precisely the answer.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (metrics tier, metrics-v1). Evidence: Default open reranker in RAG stacks; built into Open WebUI, LangChain, LlamaIndex patterns; Apache 2.0; Adds 50-300ms per query; pays off when vector hits are topically close but imprecise. License: Apache 2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Default open reranker in RAG stacks; built into Open WebUI, LangChain, LlamaIndex patterns
- · Apache 2.0
- · Adds 50-300ms per query; pays off when vector hits are topically close but imprecise