Language
Reranking
Re-order search results by true relevance.
What it is
A reranker takes a query and a list of candidate documents (for example the top 50 from a vector search) and re-orders them by how well each actually answers the query. It is a cheap, high-impact upgrade to RAG quality.
How pricing works
Usually billed per search (one query + up to N documents) or per token. We normalise to 1,000 searches of ~50 short documents.
We compare everything in one unit: USD per 1K searches.
Typical uses
- Improving RAG answer accuracy
- E-commerce and site search
- Legal/medical document retrieval
What to compare
- Relevance gain (nDCG) over plain vector search
- Max documents per call
- Latency
Watch out for
- Reranking hundreds of long documents per query gets slow and expensive; rerank only the top candidates.
All 4 APIs, cheapest first
Free to browseTick 2 to 4 APIs to compare them side by side.
| API | $ / 1K searches | Max documents per call | Multilingual | Typical latency | Quality | Price data | Compare |
|---|---|---|---|---|---|---|---|
| jina-reranker-m0 Jina AI | $0.80 | 2,048 | Yes | 400 ms | 78 | Unverified sample | |
| rerank-2.5 Voyage AI | $1.00 | 1,000 | Yes | 250 ms | 84 | Unverified sample | |
| Amazon Rerank (Bedrock) Amazon Web Services | $1.00 | 1,000 | Yes | — | – | Unverified sample | |
| Rerank 3.5 Cohere | $2.00 | 1,000 | Yes | 300 ms | 85 | Unverified sample |