You send Rerank 4 Pro a search question and a list of candidate documents, and it returns them ordered by relevance. Each document, question included, can run to about 32 pages, in more than 100 languages, and data records are accepted alongside text. Use it as a second pass after a keyword search or a search by meaning, for hard questions and long documents where result quality counts most. Rerank 4 Fast is the lighter version for quick replies; Rerank v3.5 reads only about 4 pages per document.
Rerank 4 Fast
by CohereReleased 6 April 2026
Rerank 4 Fast takes a search question and a list of candidate documents and returns them ordered by relevance, reading up to about 32 pages per document, question included, in more than 100 languages. It is the lighter of the two Rerank 4 models, suited to quick replies and heavy traffic, so use it where many searches arrive at once.
What it costs
You pay for what each request uses. No subscription, no minimum.
One search, the ordinary use
Every rate
| Meter | ↑ you send | ↓ comes back |
|---|---|---|
| Search | 0,67 DA |
per search
- When you pay
- After the answer, from your balance. Nothing is held.
- Why it varies
- One price each time.
- After each one
- The exact charge shows on the request.
Where your request runs
One company runs this model. OpenDunes sends every request to it.
You pay the price of the provider that served your request. Prices read 15:19 · speed over the last 30 minutes · uptime over 3 days.
Call it in code
One HTTP call. Any language that can send JSON can call it.
import os, requests
BASE = "https://opendunes.com/v1"
KEY = os.environ["OPENDUNES_API_KEY"]
H = {"Authorization": f"Bearer {KEY}"}
resp = requests.post(f"{BASE}/rerank", headers=H, json={
"model": "cohere/rerank-4-fast",
"query": "Is the deposit refundable if I cancel?",
"documents": [
"A cancelled order refunds the deposit within 14 days.",
"Deposits are held for the life of the account.",
"Delivery outside Algiers takes three to five days.",
],
})
print(resp.json())Runs this request once on your balance.
Questions about Rerank 4 Fast
How much does Rerank 4 Fast cost in dinars?
About 0,67 DA per search.
Is Rerank 4 Fast free? Can I try it?
No. You pay per use from your balance, with no subscription. Run it once from Call it in code, at the same price.
What is Rerank 4 Fast best for?
Rerank 4 Fast takes a search question and a list of candidate documents and returns them ordered by relevance, reading up to about 32 pages per document, question included, in more than 100 languages. It is the lighter of the two Rerank 4 models, suited to quick replies and heavy traffic, so use it where many searches arrive at once.
Rerank 4 Fast or Rerank 4 Pro?
Rerank 4 Pro costs 0,16 DA more for the same search.
How do I pay for Rerank 4 Fast without an international card?
Deposit dinars with your Edahabia or CIB card. Every model, Rerank 4 Fast included, is paid from that balance. No foreign card, no conversion.
Does Rerank 4 Fast work in Arabic, French or Darja?
Cohere says its Rerank models were trained across 100+ languages and lists Arabic and French. Darja is not named.
How much text per document?
Up to 32 768 tokens for the query and one document together.
Who makes Rerank 4 Fast, and when was it released?
Cohere. Released 6 April 2026.
Other Cohere models, priced in dinars
See all Cohere modelsCohere's earlier reranker, from before Rerank 4, reordering semantic or keyword search results by relevance and handling multi-aspect and semi-structured data across more than 100 languages
This model accepts text and pictures as input to generate numerical representations of the content. It can process up to 120 pages of material simultaneously. Its design focuses on being a fast and economical option for content representation tasks. The model is limited to processing and outputting these representations and cannot perform tasks requiring complex reasoning or generating new written material.
This model accepts text and pictures as input and outputs numerical representations of the content. It can process up to 120 pages of material at once and generate up to 110 pages of output. It is designed to be a higher accuracy tier within its family of models. This model is limited to processing and outputting these representations and cannot perform tasks requiring complex reasoning.