AI Tokens · A service of InternetONE

AI in your app,
in a few lines
of Python.

Use the OpenAI client with the AI Tokens address and key. Choose the model and send your first request.

Set up Python and the API key

Once your account is active, create your key in the dashboard. Install the client in your Python environment:

python -m pip install openai

Set the key as an environment variable. On macOS or Linux:

export AI_TOKENS_API_KEY="your-key"

On Windows PowerShell:

$env:AI_TOKENS_API_KEY="your-key"

Save the example as first.py and run python first.py in the same terminal.

Pay-as-you-go API, prices in euro, Italian invoices.
Service documentation ↗

Your first request · Python

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.ai-tokens.it/v1",
    api_key=os.environ["AI_TOKENS_API_KEY"],
)

response = client.chat.completions.create(
    model="auto",
    messages=[{"role": "user", "content": "Hello!"}],
    max_tokens=1024,
)

print(response.choices[0].message.content)

With model="auto" we pick, for each request, the model that best fits that prompt among those available at that moment: a fast one for simple tasks, one that reasons for code, calculations and planning. The model field of the answer tells you which one answered, and you pay its rates. For a specific model, write its name, as in the examples below. No request is sent from this page.

Models, machines and prices

Euro per million tokens · VAT excluded

Choose the model

Gemma 4 26B

2 hardware variants

Hardware

Inference countryItaly

Input / 1M token€0.20

Output / 1M token€0.45

Requests

Input tokens per request

Output tokens per request

Estimated total—

Tokens only, VAT excluded.

cURL

curl https://api.ai-tokens.it/v1/chat/completions

The example is a minimal request. Set AI_TOKENS_API_KEY to your key. No request is sent from this page.

API identifiergemma-4-26b@l40s

Prices from the service catalogue, updated with every change. Live availability in the dashboard.

Open the dashboard ↗

Compare all variants Full price list

18 variants in the catalogue
Model	Hardware	Inference	Input € / 1M	Output € / 1M
Apertus 70B	NVIDIA L40S	Italy	0.70	2.43
Apertus 70B	NVIDIA RTX PRO 6000 Blackwell	Italy	0.70	2.70
Gemma 4 26B	NVIDIA L40S	Italy	0.20	0.45
Gemma 4 26B	NVIDIA RTX PRO 6000 Blackwell	Switzerland	0.20	0.50
Qwen3.6 35B	NVIDIA DGX Spark (GB10)	Italy	0.20	0.66
Qwen3.6 35B	NVIDIA L40S	Italy	0.20	1.08
Qwen3-Coder 30B	NVIDIA L40S	Italy	0.12	0.45
Qwen3-Coder-Next	NVIDIA DGX Spark (GB10)	Italy	0.15	0.44
Qwen3.8 27B	AMD Radeon AI PRO R9700	Italy	0.45	2.16
Qwen3.8 27B	NVIDIA L40S	Italy	0.45	2.43
Ministral 3 14B	NVIDIA DGX Spark (GB10)	Italy	0.19	0.1045
Ministral 3 14B	Apple Mac Studio M5 Max	Italy	0.19	0.1235
Ministral 3 14B	AMD Radeon AI PRO R9700	Italy	0.19	0.152
Ministral 3 14B	NVIDIA RTX PRO 6000 Blackwell	Switzerland	0.19	0.19
Mistral Small 4	NVIDIA RTX PRO 6000 Blackwell	Switzerland	0.15	0.60
gpt-oss-120b	NVIDIA DGX Spark (GB10)	Italy	0.15	0.33
Mistral Medium 3.5	NVIDIA RTX PRO 6000 Blackwell	Switzerland	1.40	7.00
Qwen 2.5 1.5B	Hardware not specified	Switzerland	0.02	0.05

Transcription, speech synthesis, embeddings and questions on documents. See the other services ↗

A choice that matters

The machine
makes a difference too.

A model does not work in a vacuum. It runs on a machine, in a place, at a cost. We chose to show you these facts because they are part of your project.

You choose

With model@hardware you name the machine. If that variant is not available, the request is refused and nothing is charged.

Let the service choose

With the model name alone, the available hardware is used, at its rate. In the dashboard, under Usage, you see which one answered each request and at what price.

All the details in the documentation ↗

The company behind the service

InternetONE

AI Tokens is a service of InternetONE, a company working on networks, cloud and infrastructure since 1996. We make the models available through an API, so you can build them into your applications without running the machines behind them.

Questions about access or integration? Write to our team.

Write to us ↓

Before you start

How do I get access?

Create an account in the dashboard. The service is pay-as-you-go, with Italian invoices.

Can I use Lombardy's 4.0 grants?

Internet One is on Unioncamere Lombardia's list of 4.0 suppliers: in the calls that require it, consulting, training and development done with us can count as eligible costs. The page on 4.0 grants in Lombardy (in Italian) explains how it works.

Where is the data processed?

The country of a variant is where the model runs. If you limit a key to a country in the dashboard, its text requests go only to machines in that country; embeddings, rerank and voice run in Switzerland for now. Accounts, keys, logs and uploaded documents are on our servers in Switzerland. Retention and providers are described in the documentation on countries and data.

Can I use these models in my project?

Each model keeps its own licence. Check the terms of the model you choose, and test quality and performance on your own use case.

What changed recently?

Every improvement to the service is listed in the service updates, newest first.

Tell us what you are building.

A few lines are enough to start.

If you need custom software, we can build it for you on AI Tokens' models. To go into it, ask us for a free 30-minute call.

AI in your app,in a few linesof Python.