How it works Rerank 4 Pro?
Rerank 4 Pro is a Cohere ranking model for search and RAG: it reorders found snippets by relevance, works with a 32K token context window, and supports over 100 languages. No data preprocessing is required—documents are sent to the model exactly as they are in your database. In BotHub, the model is called via a unified OpenAI-compatible API: change the model name in the request and compare output with others without rewriting the integration. Access from Russia without VPN or foreign cards, pay in rubles as you go, tokens do not expire. Scenarios: improving corporate search results for internal documentation, increasing RAG assistant response accuracy, selecting relevant chunks before feeding them to a language model, parsing multilingual catalogs, and supporting research pipelines.Frequently asked questions about Rerank 4 Pro
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
It is a reranker: it does not write answers, but evaluates the relevance of a 'query-document' pair and reorders found snippets by descending relevance. Suitable for refining search and RAG systems: take top candidates from standard search and keep the ones you actually need.
It does not write finished text: the output provides relevance scores and document order, not a coherent narrative. A context of up to 32,768 tokens is available per pass, meaning you can submit long snippets and match them against the query. Output volume according to our data is up to 29,491 tokens.
There is no specific note about reasoning before answering in our data. By its role, a reranker does not reason like chat models: it immediately provides a relevance score for each 'query-document' pair. If step-by-step argumentation is needed, place a separate text model alongside it.
There is no confirmed note about function calling and structured output in our data. However, the reranker's result is an ordered list of documents with scores, which is easy to receive in code and process programmatically. A generative model in tandem is usually responsible for function calls and strict JSON.
Our data does not indicate support for images, audio, or video: work is built around text—the query and candidate documents—and scores are also returned in text format. If you need to analyze images or audio, BotHub has separate multimodal models for this.
There is no specific note about Russian language quality in our data, so we will not make promises. Matching a query and a document here does not require text generation, so you can submit Russian-language materials—but you should check the quality on your own collection and queries.