# Voyage Rerank 2.5

Voyage Rerank 2.5 is Voyage AI by MongoDB's quality-tuned generalist reranker. It accepts natural-language instructions to steer relevance scoring, supports multilingual retrieval, and handles a context window of 32K tokens.

- **Model ID:** `voyage/rerank-2.5`
- **Type:** reranking
- **Providers:** voyage
- **Context window:** 32,000
- **Maximum output tokens:** 32,000
- **Pricing:** $0.05/1M input tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/rerank-2.5

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { rerank } from 'ai';

const result = await rerank({
  model: 'voyage/rerank-2.5',
  query: 'What is the capital of France?',
  documents: [
    'Paris is the capital of France.',
    'Berlin is the capital of Germany.',
    'Madrid is the capital of Spain.',
  ],
})
```

## About

Voyage Rerank 2.5 is Voyage AI by MongoDB's generalist reranker released August 11, 2025. It reorders candidate documents returned by a first-stage retriever to lift top-k relevance, with a context window of 32K tokens, multilingual support, and instruction-following capabilities new to the rerank family.

Across 93 retrieval datasets, Voyage Rerank 2.5 improves accuracy by 7.94% over Cohere Rerank v3.5 when paired with four different first-stage methods: BM25 lexical search, OpenAI `text-embedding-3-large`, `voyage-3-large`, and `voyage-3.5`. Averaged across these first-stage retrieval methods, Voyage Rerank 2.5 outperforms Cohere Rerank v3.5, Qwen3-Reranker-8B, and `rerank-2` by 3.26%, 2.34%, and 1.35% respectively on NDCG@10.

Instruction-following lets you steer relevance scores using natural language. You can describe the kind of relevance you want ('prefer documents that cite primary sources' or 'rank legal precedents above commentary') and Voyage Rerank 2.5 adjusts its scoring accordingly. On the Massive Instructed Retrieval (MAIR) benchmark, Voyage Rerank 2.5 outperforms Cohere Rerank v3.5 by 12.70%. The context window of 32K tokens is double that of `rerank-2` and eight times that of Cohere Rerank v3.5, so longer query-document pairs fit a single rerank call without truncation.

## What to consider

Voyage Rerank 2.5 is the quality-optimized tier of the Voyage AI by MongoDB 2.5 reranker series. Pick it when reranking quality is the primary constraint and per-query cost is secondary. For workloads where latency and cost dominate, `rerank-2.5-lite` reaches similar accuracy at a lower price tier.

Use instructions to encode policy that's awkward to express in retrieval embeddings. Voyage Rerank 2.5 accepts free-form instructions such as 'prefer recent regulatory filings' or 'down-rank marketing pages' and adjusts scoring without requiring index rebuilds.

Voyage Rerank 2.5 pairs with any first-stage retriever. Voyage AI by MongoDB benchmarks it on top of BM25, OpenAI `text-embedding-3-large`, `voyage-3-large`, and `voyage-3.5`. You can keep your existing embedding stack and add Voyage Rerank 2.5 as a second-stage reranker.

## When to use

### Best For

- **High-stakes retrieval** for customer-facing search, legal discovery, and compliance where top result quality matters
- **Instruction-driven relevance** encoded in natural language without retraining embeddings
- **Multilingual reranking** across many languages in a single model
- **Long query-document pairs** that fit the context window of 32K tokens without truncation
- **RAG pipelines** that need higher top-k quality without changing the upstream embedding model

### Consider Alternatives When

- **Cost and latency dominate**; `rerank-2.5-lite` reaches similar accuracy at a lower price tier
- **You need an open-source reranker** for self-hosting; Qwen3-Reranker-8B is an option, though Voyage Rerank 2.5 outperforms it on Voyage's benchmarks
- **Your retrieval is single-language English-only** and a monolingual reranker covers the workload
- **A reranker stage may not be needed**; strong first-stage retrievers like `voyage-3.5` may meet your targets without a second pass

## Best for

- **High-stakes retrieval:** Customer-facing search, legal discovery, and compliance flows where the top result quality directly affects outcomes
- **Instruction-driven relevance:** Encode preferences such as recency, source authority, or topical focus in natural language without retraining embeddings
- **Multilingual reranking:** A single reranker covers retrieval across many languages without per-language models
- **Long query-document pairs:** The context window of 32K tokens fits lengthy filings and long documents in one rerank call
- **RAG pipelines:** Improve the top-k passed into the generator without changing the upstream embedding model

## Consider alternatives

- **Cost and latency dominate:** `rerank-2.5-lite` reaches similar accuracy at a lower price tier with reduced compute per query
- **Open-source-only constraint:** Qwen3-Reranker-8B is available for self-hosting, though Voyage Rerank 2.5 outperforms it on Voyage's published benchmarks
- **Single-language English-only retrieval:** A monolingual reranker may suffice when no other languages are in scope
- **No reranker stage needed:** Strong first-stage retrievers like `voyage-3.5` may meet your accuracy targets without a second pass

## Frequently asked questions

### What is instruction-following in Voyage Rerank 2.5?

Instruction-following lets you steer relevance scores with natural language. You pass an instruction along with the query and candidate documents, and Voyage Rerank 2.5 adjusts scoring to reflect the instruction. On the MAIR benchmark, this capability lifts Voyage Rerank 2.5 12.70% above Cohere Rerank v3.5.

### What is the difference between Voyage Rerank 2.5 and rerank-2.5-lite?

Voyage Rerank 2.5 is the quality-optimized tier and `rerank-2.5-lite` is the latency-and-quality balanced tier. Both support instruction-following and the same context window of 32K tokens. Choose Voyage Rerank 2.5 when top-result accuracy is the priority; choose `rerank-2.5-lite` when per-query cost or latency matters more.

### How does Voyage Rerank 2.5 compare to Cohere Rerank v3.5?

Across 93 retrieval datasets, Voyage Rerank 2.5 improves accuracy by 7.94% over Cohere Rerank v3.5. Averaged across four first-stage retrieval methods, it outperforms Cohere Rerank v3.5 by 3.26% on NDCG@10 and by 12.70% on the MAIR instruction-following benchmark.

### What is the context window for Voyage Rerank 2.5?

32K tokens. That is double the window of `rerank-2` and eight times that of Cohere Rerank v3.5, so longer query-document pairs fit a single rerank call without truncation.

### Which first-stage retrievers does Voyage Rerank 2.5 work with?

Voyage Rerank 2.5 reorders candidates from any first-stage method. Voyage AI by MongoDB benchmarks it on BM25 lexical search, OpenAI `text-embedding-3-large`, `voyage-3-large`, and `voyage-3.5`. You can keep your existing embedding model and add Voyage Rerank 2.5 as a second-stage reranker.

### Does Voyage Rerank 2.5 support multilingual retrieval?

Yes. Voyage Rerank 2.5 reranks across many languages without separate per-language models. Voyage AI by MongoDB reports consistent improvement across all evaluated languages and first-stage retrieval methods.

### How do I access Voyage Rerank 2.5 through Vercel AI Gateway?

Add your Voyage AI by MongoDB API key in AI Gateway settings, then send rerank requests through AI Gateway. AI Gateway authenticates requests and records usage. You can call Voyage Rerank 2.5 through the AI SDK alongside Chat Completions, Responses, and Messages API formats.

### Is Zero Data Retention available for Voyage Rerank 2.5?

Zero Data Retention is not currently available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

## Links

- [Model page](https://vercel.com/ai-gateway/models/rerank-2.5)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://docs.voyageai.com/docs/reranker)
- [Provider pricing](https://docs.voyageai.com/docs/pricing)


If you encounter incorrect information, can’t find what you need, need a capability Vercel doesn’t currently offer, or have any other feedback to report, please do it by following the instructions at https://vercel.com/feedback.md.