We use cookies. Cookies and browser storage keep you signed in and remember your preferences. With your permission we also measure how the site is used. You can any time. See our cookie policy.
mongodb/rerank-3Runs on your balance at the prices above. New accounts start with free usage.
Settings
VoyageAI by MongoDB: rerank-3 is a reranking model from VoyageAI by MongoDB. It sits in the rerank category and is positioned as a drop-in upgrade to rerank-2.5, which means teams already using the earlier model can move to rerank-3 without reworking their integration. The model is built for retrieval quality rather than generation, so it belongs in the part of a pipeline that decides how search results or candidate passages should be ordered before they reach a user or a downstream model.
The model takes text as input and produces a reranking output. It supports a context length of 32,000 tokens, with a maximum output of 28,800, which is enough room to score long candidate passages alongside a query rather than only short snippets. Because the input modality is text and the output modality is rerank, the model does not generate prose or chat replies; it returns an ordering of the passages it is given. That makes it a scoring and sorting component rather than a conversational one.
Practical use cases follow directly from that shape. A retrieval pipeline can fetch a broad set of candidates with a fast first-stage search, then pass the query and those candidates to rerank-3 to produce a better final ordering. This is useful in search over documentation, internal knowledge bases, support content, or any corpus where the top results matter more than raw recall. It also fits retrieval-augmented generation setups, where the quality of the passages placed into a prompt affects the answer that follows. The long context and long output allowance make it suitable for collections that include lengthy documents, since the model can consider more of each candidate when assigning its position.
You call the model through api14's OpenAI-compatible API using the model id "mongodb/rerank-3". Because the interface follows the OpenAI format, you can point an existing OpenAI client at api14 and select this model id in your request. The model lists no supported parameters beyond the model id itself, so you should not expect to tune sampling or generation settings such as temperature or top_p; those belong to generative models, not to a reranker. Send the query and the candidate texts, and read back the reranking output.
It is worth noting that rerank-3 is a reranker, so it is not the right choice when you need text generation, chat, summarization, or embeddings. If your task is to produce an answer rather than to order candidates, a generative model fits better. If your task is to create the initial vector representations that a reranker then reorders, an embedding model is the appropriate component. Within reranking work, rerank-3 is the drop-in successor to rerank-2.5, and the provider reports improvements over that model of 0.80% NDCG@10 on average across domain evaluations and 3.35% on long-document evaluations.
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api14.com/api/v1",
apiKey: process.env.API14_API_KEY,
});
const result = await client.post("/rerank", { body: {
"model": "mongodb/rerank-3",
"query": "What is the capital of France?",
"documents": [
"Paris is the capital of France.",
"Berlin is in Germany."
],
"top_n": 1
} });
console.log(result.results);
This community has no topics yet. Open the first one about VoyageAI by MongoDB: rerank-3.