Voyage AIVoyage Rerank 1
rerank-lite-1
Rerank Lite 1 introduced Voyage's reranking service. It uses a cross-encoder to score the query together with each candidate document.
Model design
The release describes using the reranker after either keyword retrieval or embedding search. Its launch evaluates 27 retrieval datasets.Voyage introduces rerank-lite-1
Practical use
Practical guidance. This two-stage pattern can improve a keyword search without first replacing the whole index. Measure the candidate stage and the final ranking separately. A useful candidate that falls outside the supplied set remains invisible to the reranker.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API |
| API limits | 4,000 tokens per pair; 1,000 per queryVoyage AI reranking API and earlier models |
| Access | Earlier Voyage API catalogVoyage AI reranking API and earlier models |
Related models
- rerank-2-lite
Voyage AI · Voyage Rerank 2
Smaller second-generation reranker with an 8,000-token pair limit.
Launch date API Earlier general and multilingual relevance reranker.
Launch date API