Skip to main content


At the heart of our payment system is a pay-per-use policy. We will neither charge you on a regular basis, nor can you ever have a negative balance with us – Instead, you have to top up your account with so-called credits. The price of five credits is one Euro.

Each request is priced the same; whether it's send via the playground or API. Each time you send a request, your input is translated into tokens – the AI’s language if you will. The cost of each request is solely based on the number of tokens the model receives and generates. If you use the parameter "Best of", your request will be executed as often as the set value.

Pricing Example: Text-only

For clarity, let's go over a pricing example. This article on Aleph Alpha is approximately 580 words (about 3800 characters without spaces or one DIN A4 page) long. We now want to analyze this text using some model, so we send it, plus some natural language prompt, to the API.


After stripping the Heise article of external content and annotations, we copied it to a Microsoft Word file. In 11pt Times New Roman with a line spacing of one, the article occupies almost exactly one DIN A4 page. Therefore, this pricing example can serve as a benchmark for calculating the cost of processing documents based on the number of pages.

The following prompt setup was used:

Text: "[vollständiger Artikeltext]"
Q: Wie heißt die Aleph Alpha Funktionalität, die auch Bilder versteht?

Or, in English:

Text: "[entire article text]"
Q: What is the name of the Aleph Alpha functionality that also understands images?

Either way, the AI will correctly answer:


The token length of the prompt (including the article) and completion are 917 and 2, respectively. Thus, the cost of the (German) request for each model will be:

ModelCost (Credits)

Pricing Example: Multimodal

You may have noticed that the Heise article featured a schematic about how MAGMA works. If we wanted to include that image into our prompt to process the article in its entirety, the costs change. Our prompt now looks like this:

[beginning article text]

MAGMA Function
[image caption]
[end article text]
Q: What animal is in the picture?
A:  Goose

Notice how the model can be effortlessly prompted in any language. This latest request includes 926 text tokens and one image, which if put in a word document (see "Pricing Example: Text-only" for fromatting information) measure up to almost exactly 1,5 pages. From the number of tokens, we can calculate the cost for this multimodal request.

ModelCost (Credits)