Voyage Rerank 2.5 is Voyage AI's generalist reranker optimized for retrieval quality. It supports a context window of 32K tokens, instruction-following over natural language steering, and multilingual reranking. On 93 retrieval datasets, it improves accuracy by 7.94% over Cohere Rerank v3.5.
On-demand DeploymentDocs | On-demand deployments allow you to use Voyage ReRank 2.5 on dedicated GPUs with Fireworks' high-performance serving stack with high reliability and no rate limits. |
Instruction-following rerankers allow users to guide how relevance should be evaluated using natural language instructions.
For example, a legal search application could instruct the model to prioritize regulatory documents over court cases. A finance application could instruct the model to rank disclosure from the modest recent fiscal quarter above older filings. These instructions help the reranker better align results with user intent.
Use rerank-2.5 when retrieval accuracy is a priority and you want to improve the ranking of results returned by a first-stage search system.
It is especially useful for RAG applications, enterprise search, legal and financial research, technical documentation, medical search, code search, multilingual retrieval, and long-document retrieval.
Embedding models are typically used for first-stage retrieval, where they quickly find a set of potentially relevant documents. rerank-2.5 is then applied as a second-stage model to score and reorder those candidates.
This two-stage approach combines efficient retrieval with higher final ranking accuracy.
