rerank-3
Rerank 3 updates Voyage's reranking backbone and training mixture. It ranks supplied documents against a query and can follow ranking instructions.
Model design
The release emphasizes long-document and code retrieval. Voyage calibrated scores toward the corresponding 2.5 model's distribution to ease migration. Its evaluation reranks a retrieved candidate set, rather than searching an entire corpus directly.rerank-3 and rerank-3-lite
Practical use
Practical guidance. A reranker cannot recover a source that the first retrieval stage missed. Check candidate recall before increasing reranker size. Recheck any score threshold on real queries even when the provider aims to preserve score distributions.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API |
| Context window | 32,000 tokens Query and one document together. Voyage AI reranking API and earlier models |
| Atlas active date | 29 September 2026MongoDB Voyage AI model lifecycle: Atlas scope and release history |
| Query limit | 8,000 tokensVoyage AI reranking API and earlier models |
Related models
- rerank-2.5
Voyage AI · Voyage Rerank 2.5
Reranker that follows natural-language ranking criteria.
Launch date API - rerank-3-lite
Voyage AI · Voyage Rerank 3
Smaller instruction-following reranker for interactive retrieval.
Launch date API