bge-reranker-large

Large BGE reranker for rescoring retrieved passages after vector search in search and RAG pipelines.

BAAI · BGE

Editorial review

Reviewed byOpenSourcesAI EditorialLast updatedAugust 2026SourcesExact model card

Model checkpoints, context windows, provider support, local runtime compatibility, and license terms can change quickly. Verify the exact model card before production or commercial use.

Best for

Teams that already retrieve a candidate set and want a stronger final ranking step before passing context to an LLM.

Who should use it

Teams that already retrieve a candidate set and want a stronger final ranking step before passing context to an LLM.
Builders who want local or self-hosted testing options.
RAG builders comparing retrieval quality, latency, and multilingual coverage.

Common workflows

RAG reranking workflows
reranking workflows
rag workflows
retrieval workflows
semantic search workflows

Deployment and hardware notes

~2.24 GB in fp32 (560M parameters), ~1.12 GB in fp16 — plan on a GPU for interactive reranking beyond small candidate lists.

License and usage notes

MIT. Open weights. Verify the exact model card and license terms for the checkpoint or hosted provider you use.

Strengths

Open weights model option for BGE workflows.
Teams that already retrieve a candidate set and want a stronger final ranking step before passing context to an LLM.
A drop-in upgrade for bge-reranker-base pipelines: same 512-token pair format, same score-only output, roughly double the compute per pair.

Limitations

Twice bge-reranker-base's weight for measurably better pair ranking; the same per-query cost scaling applies, so candidate-list size drives latency. Trained mainly on Chinese and English pairs.
~2.24 GB in fp32 (560M parameters), ~1.12 GB in fp16 — plan on a GPU for interactive reranking beyond small candidate lists.
Context window and limits: 512 tokens.
Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

A drop-in upgrade for bge-reranker-base pipelines: same 512-token pair format, same score-only output, roughly double the compute per pair.

Local runtimes: Ollama where supported, LM Studio where supported, llama.cpp where supported, Transformers

Platforms: Windows, macOS, Linux

Reranker spec

Memory~2.24 GB in fp32 (560M parameters) · ~1.12 GB in fp16OutputRelevance score per query-passage pair (a cross-encoder emits no vectors)Max input512 tokens

Outputs a relevance score per (query, passage) pair — no vectors, so nothing new is indexed. Prefer bge-reranker-v2-m3 when candidates are multilingual or longer than 512 tokens.