by Voyageai·Released 30 September 2026
This model takes in written text only and gives back the same candidates reordered, most relevant first. Its work is the second pass of a search, where the first results are sorted again so the best matches sit on top, and it reads about 32 pages of text at once. It is tuned for quick turnaround on busy pages and is a little more accurate than the earlier lite version. It cannot write new text or read pictures, so a job that needs a written answer calls for a different kind of model.
You pay for what each request uses. No subscription, no minimum.
One search, the ordinary use
Every rate
| Meter | ↑ you send | ↓ comes back |
|---|---|---|
| Text tokens | 6,66 DA |
per 1M tokens
One company runs this model. OpenDunes sends every request to it. DA per 1M tokens: ↑ what you send.
| Provider | Latency | Uptime |
|---|---|---|
| vVoyageAI by MongoDB | … | … |
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.
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": "voyageai/rerank-3-lite",
"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.
About 0,0067 DA per search. ↑ 6,66 DA per 1M tokens.
No. You pay per use from your balance, with no subscription. Run it once from Call it in code, at the same price.
This model takes in written text only and gives back the same candidates reordered, most relevant first. Its work is the second pass of a search, where the first results are sorted again so the best matches sit on top, and it reads about 32 pages of text at once. It is tuned for quick turnaround on busy pages and is a little more accurate than the earlier lite version. It cannot write new text or read pictures, so a job that needs a written answer calls for a different kind of model.
rerank-3 costs 0,010 DA more for the same search.
Deposit dinars with your Edahabia or CIB card. Every model, rerank-3-lite included, is paid from that balance. No foreign card, no conversion.
Voyageai names no languages for rerank-3-lite. Arabic, French and Darja: not stated.
Up to 32 000 tokens for the query and one document together.
Voyageai. Released 30 September 2026.
Reorders search candidates by relevance as the quality-tuned reranker that upgrades rerank-2.5, gaining most on long documents and code, scoring each query and document pair up to 32K tokens and following instructions on what counts as relevant
Voyage-code-4 turns computer code into a fingerprint of its meaning, so a search finds the code that does what you describe, not just code with the same words. Voyage built it for coding assistants and code search, and it reads about 30 pages at a time. Pick it to search through a code base. For ordinary writing, voyage-4 is the general-purpose model.
Rerank 2.5 Lite takes a question and a list of candidate documents and puts them in order of how well they answer it, which sharpens the results of a search. You can add written instructions to steer it, and a question and one document together can run to about 30 pages. It is tuned for speed.
Rerank 2.5 takes a question and a list of candidate documents and puts them in order of how well they answer it, which sharpens the results of a search. You can add written instructions to steer it, such as which kind of document to favour. A question and one document together can run to about 30 pages.
From what its maker and two benchmark sites publish, and where it stands on OpenDunes. Read 3 Nov 2026.
Sorting a long list of search results so the best hits come first
Ordering a batch of long candidate documents against one query at a time