What does a million tokens cost?

These are AI models you rent from six European providers: OVHcloud, Scaleway, Infomaniak, Mistral AI, IONOS and STACKIT. This page is a parts list: click up to four models to compare them, or open the "Same model" tab to see where each model is cheapest.

A token is a small chunk of text, about three quarters of a word. Providers charge per million tokens. A million tokens is roughly 750,000 words. Input tokens are what you send; output tokens are what the model writes back.

How renting works

Your software sends text to an endpoint, the provider's web address for the model, and the model sends back an answer. You sign up with a provider, create an API key that works like a password, and pay for what you use.

Prices are shown in euros; switch to pounds or dollars at the top of the page. We convert from each provider's own billing currency, so the figure on your invoice may differ a little.

What software to use
ContextWhat you give it: your request, your files, your rules.
App (harness)The software around it: Cherry Studio, TypingMind, n8n, OpenCode.
ModelThe engine underneath: the 75 models on this page, from six European providers.
This page prices the bottom layer. You install the middle layer; you supply the top layer.

The software side is easier than it sounds. Every tool asks for the same three values: a base URL, an API key and a model ID. The base URL is the provider's address; the API key works like a password; the model ID names the model. All six providers here use addresses ending in /v1; copy yours exactly from the provider's page.

Start with the provider's own playground, which needs no extra software. Add a chat app when you want to talk to a model day to day. Move to an automation tool when the model should work inside a workflow. Take on a coding agent when it should write and check code.

Cherry Studio is a free, open-source chat app for your desktop. TypingMind is a paid chat app in your browser, with a one-off licence. n8n is a visual automation builder; its cloud runs in the EU. OpenCode is a free coding agent; IONOS and STACKIT, two providers here, publish setup guides for it.

Open WebUI and LibreChat do a similar job, but you host them yourself. They suit people who already run servers.

Editor's picks

Where to start

If you just want a sensible answer, start with these. Click a card to add it to the comparison tray. Monthly figures assume the Compare tab's default: 7.5 million words in and 1.5 million back, about 10M tokens in and 2M out.

Free to try: OVHcloud's playground needs no account, and it runs five small models at no cost. Scaleway gives a million tokens, about 750,000 words, a month. Infomaniak gives a million credits for your first month. Mistral ended its free option for its API in October 2026; its API now needs a payment method.

What these words mean

A lab builds a model; a provider runs it and sends the bill. The card names the provider; "More details" names the lab. An endpoint is the web address your software sends requests to.

Token
A small chunk of text, about three quarters of a word. Providers charge per million tokens.
Input and output
Input is the text you send. Output is the text the model writes back.
Context
How much text the model can work with in one go. 128K tokens is about 96,000 words. The diagram above uses the same word for what you give the model.
App (harness)
The software you run the model in: Cherry Studio, TypingMind, n8n, OpenCode. The diagram above uses two words for it; app and harness mean the same thing here.
Parameters
A rough measure of a model's size. Bigger usually means more capable and more expensive. B means billion.
Name codes
The numbers in a model's name describe its size: 397B means 397 billion parameters; A17B means about 17 billion are active at once. FP8 is a compact number format that makes a model cheaper to run.
Mixture of experts
A model that consults only a small share of itself for each word, so it runs fast and cheap for its size.
Open weights
The model files are public, so anyone may run it themselves. Here you rent it from a provider. Closed weights stay with the provider.
Combined price
Input plus output, each per million tokens. A quick way to rank models; the Compare tab works out a realistic monthly bill.
Embeddings
Models that turn text into numbers, so software can search by meaning.
Audio
Models that turn recorded speech into text, billed by the hour of audio.
Agent
A model set up to finish a task in steps, using other software as it goes.
Licence
The rules for what you may do with the model. Permissive licences, such as Apache 2.0, allow almost any use, including paid products. Restricted licences add conditions.
Credits
Infomaniak's own unit of free use. Its allowance is one million credits for the first month.
Free tier
An allowance you can spend before paying anything. Scaleway gives a million tokens a month; Infomaniak gives a million credits for your first month. Mistral's free tier no longer includes the API (changed October 2026).
Base URL
The provider's web address for its models. Every provider here uses one ending in /v1.
API key
A password your software shows the provider so it knows who is paying. You create it in your provider's account.
Model ID
The name of a model in a provider's list, such as Qwen3.8-27B. The software uses it to pick the model.
Must support:
Combined price per 1M tokens: up to $0.60 $0.60 to $2.00 over $2.00 | Model size (billions of parameters): up to 10B 10B to 70B 70B or more