Rerank model comparison

Four families dominate the reranking space. Here's how they compare on quality, speed, language support, and cost — so you can pick the right one for your stack.

Model Type Best for Languages Pricing BEIR NDCG@10
bge-reranker-v2-m3 Open weights Self-hosted, free, multilingual 100+ langs Free (self-host) ~60.1
Cohere Rerank v3.5 Hosted API Best multilingual quality, mature API 100+ langs ~$0.002/1k docs ~61.2
Jina Reranker v2 Open + Hosted Open weights + API + tiny browser model 100+ langs Free tier + pay-as-you-go ~59.6
Voyage Rerank 2 Hosted API High retrieval precision, domain-specific Multilingual ~$0.0005/1k docs ~61.0

BEIR NDCG@10 figures are approximate and vary by dataset — treat them as a rough guide. Check each vendor's benchmarks for task-specific results.

How to choose

Use bge-reranker if…

  • You want to self-host with no per-call cost
  • You need strong multilingual reranking for free
  • You're already using sentence-transformers and want a drop-in

Use Cohere Rerank if…

  • You want a battle-tested hosted API with official SDK support
  • Multilingual quality is the top priority
  • You're already in the Cohere ecosystem

Use Jina Reranker if…

  • You want the option to self-host OR call an API
  • You need a tiny model that can run in the browser or on edge
  • You want a free tier with no credit card required

Use Voyage Rerank if…

  • You need top-tier retrieval-specific precision
  • You're working on code search or financial documents
  • You're already using Voyage embeddings

Try a cross-encoder live

See how any of these models would reorder your retrieval results — demo runs in your browser.

Open the demo →