# Munito AI is live: open-model inference, run in Europe

URL: https://munito.ai/blog/launching-inference

2026-10-23

One endpoint for DeepSeek, GLM, Qwen and more, on GPUs we rent and run. Keep your OpenAI and Anthropic libraries, pin the region, and pay a posted price.

## Munito AI is open to everyone

Munito AI is an inference provider for Europe. One endpoint serves the open models worth running, on GPUs we rent and operate: DeepSeek, GLM, Qwen and more, for chat, tool calling and structured outputs, plus OCR that reads a PDF or a scan by the page.

The private beta is over. Sign up, take an API key, and send a request. See [the models we serve](https://docs.munito.ai/models) and the [Quickstart](https://docs.munito.ai/quickstart).

## Keep the SDK you already use

The API answers in two shapes on one key. An OpenAI-shaped client changes a base URL. An Anthropic-shaped client, Claude Code included, changes a base URL. No proxy sits in between, and tool calling, streaming and structured outputs work as they do today.

```python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.eu.munito.ai/inference/v1",
    api_key="munitoai_...",
)
```

The same account answers the Anthropic Messages API at `https://api.eu.munito.ai/inference`. See [Compatibility](https://docs.munito.ai/compatibility).

## Pin the region, and prove it

Send to `https://api.eu.munito.ai` and the request runs in the EU. The response names the region in the `munito-region` header, so the proof is in the answer rather than in a brochure. A US host and a global host answer the same way. See [Regions](https://docs.munito.ai/regions).

We keep no prompts and no outputs by default, we train on nothing you send, and no person reads your content. The DPA applies with no signature, at [legal.munito.ai/dpa](https://legal.munito.ai/dpa), and the [sub-processor list](https://trust.munito.ai/subprocessors) names every party of each region.

## We own no GPUs, and that is the point

Munito AI rents capacity across clouds and GPU markets, places the replicas, runs the model servers and fronts each model with a router of its own. A launch waits on no purchase order, and a region is a choice rather than a building we had to finance.

What you see is the result: a posted price for each model, capacity that grows with the work, and a move from shared capacity to GPUs reserved for you when your traffic earns it.

## What it costs

Every price is published, per million tokens or per page, with a free tier to start and no monthly floor. See [Pricing](https://docs.munito.ai/pricing) and [the prices of each model](https://docs.munito.ai/models).

## Where it runs next

Managed in the EU or the US today, inside your own cloud account or on-premises when your rules ask for it, with one API across the three. See [Inference](https://munito.ai/inference) and [Enterprise](https://munito.ai/enterprise), or [talk to an engineer](https://munito.ai/contact).