opendunes

voyage-4-large

by MongodbReleased 27 July 2026

This model turns written text into long lists of numbers that a search system can compare, so it is built for finding the right passage in a large pile of documents rather than for writing prose. It reads up to about 32 pages in one go and handles text in many languages, which most models of this kind do not. Its number lists can be shortened to save storage space without losing much of the match quality, a trick few of its kind offer. It gives back no words at all, so it cannot draft, summarise or answer questions on its own; that work needs a model that writes text.

TakesText
GivesEmbeddings
Create API keyCompare

What it costs

You pay for what each request uses. No subscription, no minimum.

≈ 0,040 DA per page3 000 DA buys about 75 000 pages

One page, the ordinary use

1 000 tokens sent× 39,94 DA per 1M0,040 DA
Total≈ 0,040 DA

Every rate

Meter↑ you send↓ comes back
Text tokens39,94 DA

per 1M tokens

Build a bilingual search index from 50 pages of French and English text≈ 2,00 DA
Embed 200 customer support tickets for retrieval≈ 2,40 DA
Index a 300-page handbook for semantic search≈ 11,98 DA
When you pay
After the answer, from your balance. Nothing is held.
Why it varies
The length of the text.
After each one
The exact charge shows on the request.
DepositDeposits from 50 DA, with Edahabia or CIB.

Where your request runs

One company runs this model. OpenDunes sends every request to it. DA per 1M tokens: ↑ what you send.

VoyageAI by MongoDB…
↑ text
39,94 DA
Latency
…
Provider↑ textLatencyUptime
VoyageAI by MongoDB39,94 DA……

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

Swap the base URL and the key. Your OpenAI code runs unchanged.

POST /v1/embeddings
Base URL
https://opendunes.com/v1
Model ID
mongodb/voyage-4-large
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://opendunes.com/v1",
    api_key=os.environ["OPENDUNES_API_KEY"],
)

resp = client.embeddings.create(
    model="mongodb/voyage-4-large",
    input="Delivery to Oran takes two days.",
)

print(resp.data[0].embedding[:5])
pip install openaiFull reference for this route
Sign in to run it

Runs this request once on your balance.

Questions about voyage-4-large

  • How much does voyage-4-large cost in dinars?

    About 0,040 DA per page. ↑ 39,94 DA per 1M tokens.

  • Is voyage-4-large 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 voyage-4-large best for?

    This model turns written text into long lists of numbers that a search system can compare, so it is built for finding the right passage in a large pile of documents rather than for writing prose. It reads up to about 32 pages in one go and handles text in many languages, which most models of this kind do not. Its number lists can be shortened to save storage space without losing much of the match quality, a trick few of its kind offer. It gives back no words at all, so it cannot draft, summarise or answer questions on its own; that work needs a model that writes text.

  • voyage-4-large or voyage-4-lite?

    voyage-4-lite costs 0,033 DA less for the same page.

  • How do I pay for voyage-4-large without an international card?

    Deposit dinars with your Edahabia or CIB card. Every model, voyage-4-large included, is paid from that balance. No foreign card, no conversion.

  • Does voyage-4-large work in Arabic, French or Darja?

    Mongodb names no languages for voyage-4-large. Arabic, French and Darja: not stated.

  • How much text in one input?

    Up to 32 000 tokens. Split anything longer.

  • How many dimensions?

    Not stated.

  • Who makes voyage-4-large, and when was it released?

    Mongodb. Released 27 July 2026.

Other Mongodb models, priced in dinars

See all Mongodb models
voyage-4Mongodb · Jul 27, 2026

This model turns written text into compact number lists that a search system can compare, so it is built for finding the right passage in a large pile of documents rather than for writing prose. It reads up to about 32 pages in one go and can handle text in many languages, which most text models of this kind do not. It gives back no words at all, so it cannot draft, summarise or answer questions on its own; that work needs a model that writes text.

32 pages per call
≈0,020 DAper page
View model
voyage-4-liteMongodb · Jul 27, 2026

This model turns written text into compact number lists that a search system can compare, so it is built for finding the right passage in a large pile of documents rather than for writing prose. It reads up to about 32 pages in one go and gives back no words at all, so it cannot draft, summarise or answer questions on its own; that work needs a model that writes text. It is the lightest of its family, tuned to answer quickly and cheaply, which makes it a fit when you have a great many passages to sort through and speed matters more than the finest match.

32 pages per call
≈0,0067 DAper page
View model
voyage-code-4Mongodb · Aug 13, 2026

voyage-code-4 is a code embedding model from Voyage AI, a MongoDB company. It is designed for coding agents and code retrieval, with Matryoshka embeddings at 2048t voyage-code-4 here: [mongodb.com/products/platform/ai-search-and-retrieval/models](https://www.mongodb.com/products/platform/ai-search-and-retrieval/models?utm_source=OpenDunes&utm_medium=referral)

32 pages per call
≈0,040 DAper page
View model
voyage-multimodal-3.5Mongodb · Jul 27, 2026

This model takes in written text and pictures, reading up to about 32 pages at once. For whatever it reads, it gives back a numeric fingerprint that can stand for as much as 28 pages, and software compares those fingerprints to find matches or rank results by closeness. Its work is letting a buyer search a mixed archive of reports, slide decks and photos for the pages that fit a question, and it stops at audio and video, which it does not accept, so those files need a different kind of model.

32 pages per call
≈0,040 DAper page
View model