VoyageAI by MongoDB: rerank-2.5

VoyageAI by MongoDB: rerank-2.5 is a reranking model from Voyage AI, a provider whose rerankers are distributed under the VoyageAI by MongoDB name. It sits in the rerank category, which means its job is not to generate text but to take a set of candidate documents and reorder them by relevance to a query. If your application already retrieves passages, chunks, or search results and you want the best matches at the top, this is the kind of model you would reach for.

The model accepts text as input and produces a reranking as output. It has a context length of 32,000 tokens, which is the budget shared across the query and the documents you submit in a single request, and a maximum output of 28,800. Because the output is a ranking rather than generated prose, you should not expect it to answer questions, summarize content, or write code. Its entire contribution is a relevance ordering over the candidates you provide.

Practical use cases follow directly from that. A retrieval augmented generation pipeline can use rerank-2.5 as a second stage after a fast vector or keyword search, taking the top candidates and reordering them so the most relevant chunks are the ones passed to a generation model. The same pattern applies to enterprise search, documentation search, and support knowledge bases, where the first stage favors recall and the reranker restores precision. It is also useful for deduplication and prioritization tasks where you have more candidate passages than you can afford to show a user or feed into a downstream prompt.

You call it through api14's OpenAI-compatible API using the model id "voyageai/rerank-2.5". Because the endpoint follows the OpenAI schema, you can point an existing OpenAI client at api14's base URL and pass this id in the model field. The model's supported parameters list is empty, so there are no additional tuning options to set beyond what the rerank request itself requires, such as the query and the documents to be scored. Keep requests within the 32,000 token context, remembering that the query and every candidate document count toward that total.

On the question of when to look elsewhere: rerank-2.5 only handles text and only returns a ranking, so tasks that need image input, audio, or generated output are outside its scope and belong to a different model. If you need a single model to both retrieve and answer, a generation model is the better fit, with rerank-2.5 optionally placed in front of it as a precision stage.