Cohere: Rerank 4 Fast

Cohere: Rerank 4 Fast is a reranking model from Cohere, released by the provider on 6 April 2026. It sits in the rerank category rather than the chat or generation category: instead of writing an answer, it scores and reorders a list of candidate documents against a query so that the most relevant items appear first. Cohere describes it as an AI search foundation model built to improve the relevance of information surfaced within search and retrieval augmented generation systems.

The model takes text as its only input modality and produces rerank output, meaning an ordered list of documents with relevance scores. It supports a context window of 32,768 tokens, which applies across the query and the documents passed in a single request, and it has a maximum output of 29,491. Multilingual support covers more than 100 languages, and the provider states that no data pre-processing is required before you send text to the model.

Those capabilities point to a fairly specific set of uses. The most common is a second stage in a retrieval pipeline: a fast first-stage retriever returns a broad set of candidates, and Rerank 4 Fast reorders them so the passages most likely to answer the query land at the top before they reach a generation model. The same pattern applies to enterprise search, where relevance ordering matters more than generating a reply, and to RAG systems that need to trim noisy context before prompting an LLM. Multilingual support makes it usable in pipelines that handle queries and documents in many languages at once, and the 32K context allows a long query plus a substantial batch of candidate passages in one call.

You call it through api14 using the OpenAI-compatible API with the model id "cohere/rerank-4-fast". Because api14 exposes an OpenAI-compatible surface, you can point your existing OpenAI client at the api14 base URL and change the model field to that id, rather than adopting a separate SDK. The model accepts text input and returns rerank output. No additional request parameters are listed as supported for this model, so plan your integration around the defaults and the prompt and document content you send, and check the live price table on the page for current rates.

One practical constraint to keep in mind is the 32K context window, which has to cover the query and all candidate documents in a request. If your documents are long or you need to rank a very large candidate set in one pass, split the work into batches or shorten the passages first. Rerank 4 Fast also will not generate text; if you need summarisation, question answering or any conversational output, pair it with a generation model and use this one purely for ordering.