---
title: "Magistral Small API hosted in the EU"
description: "Run Magistral Small 1.2, Mistral AI's open 24B reasoning model, on inference.eu through an OpenAI-compatible API with processing in EU data centres."
canonical: "https://inference.eu/models/magistral-small/"
language: "en"
---

Reasoning · Mistral AI

# Run Magistral Small in the EU through an OpenAI-compatible API

Magistral Small is the open reasoning model from Paris-based Mistral AI: a 24-billion-parameter model that thinks through a problem before it answers, and since version 1.2 also reads images. On inference.eu you call it with the OpenAI or Anthropic request format, processed in EU data centres and never used for training.

[Contact sales](https://inference.eu/contact/)[Example request](#example)

1.  0124B
2.  02EU data centres
3.  03Apache 2.0

## What Magistral Small is

Magistral Small 1.2, published as Magistral-Small-2509 in September 2025, is built on Mistral Small 3.2 and trained further for reasoning, first with supervised fine-tuning on traces from the larger Magistral Medium and then with reinforcement learning. It is a dense model with 24 billion parameters.

Before answering, the model writes out its reasoning between dedicated thinking tags, then gives a final answer. Version 1.2 adds a vision encoder, so it can reason about images as well as text. The context window is 128K tokens, although Mistral notes that performance may degrade beyond about 40K. The weights are released under Apache 2.0.

## What it does well, and where it falls short

Magistral Small is good at problems that benefit from step-by-step thinking: maths, logic, planning, checking calculations and explaining a decision. Its reasoning trace is readable, which helps when you need to show how an answer was reached. It supports dozens of languages, including French, German, Italian, Spanish, Polish and Portuguese, and its size keeps it relatively light to serve.

Thinking costs tokens and time, so it is the wrong choice for simple chat or extraction. Very long inputs are a weak spot, given the drop-off past roughly 40K tokens. Mistral Small 4 now combines reasoning, vision and coding in one larger model, and is often the better default for new projects.

## Calling Magistral Small through the API

inference.eu uses the OpenAI request format, so the OpenAI SDKs, LangChain, LlamaIndex and other compatible tools work unchanged. Set the base URL to the inference.eu API, use your inference.eu key and set the model to `magistral-small`.

Mistral recommends using the model's default system prompt, which tells it to think before answering, and sampling with `temperature` 0.7 and `top_p` 0.95. Leave enough room in `max_tokens` for the reasoning trace as well as the answer. Images go in as `image_url` content parts. The Anthropic Messages format is also available for tools built around Claude.

## A French reasoning model, processed in the EU

Magistral Small is built by Mistral AI in France, and on inference.eu it runs in the European Union too. Your prompts, images, the model's reasoning trace and its answers are processed in EU data centres, and none of it is used to train any model.

We sign a data processing agreement under Article 28 of the GDPR with every customer and share our subprocessor list. Billing is in euros. The Apache 2.0 licence allows commercial use without user or revenue limits, so you can use the model in paid products, internal tools and customer-facing services without a separate agreement with Mistral.

## Magistral Small 1.2

Maker

Mistral AI

Made in

France

Type

Reasoning

Parameters

24B

Context window

131,072 tokens

Licence

[Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)

Released

September 2025

Model ID

magistral-small

Processing

EU data centres

Training on your data

Never

Source

[huggingface.co](https://huggingface.co/mistralai/Magistral-Small-2509)

magistral-small · Python

```
import os
from openai import OpenAI

client = OpenAI(base_url="https://api.inference.eu/v1", api_key=os.environ["INFERENCE_EU_API_KEY"])

response = client.chat.completions.create(
    model="magistral-small",
    messages=[{"role": "user", "content": "Summarise this contract in five bullet points."}],
)
print(response.choices[0].message.content)
```

## What teams use it for

-   1

    ### Calculation and plan checking

    Check quotes, schedules and calculations step by step and point out where a figure or assumption does not add up.

-   2

    ### Explainable decisions

    Produce a written line of reasoning alongside an answer, for reviewers who need to see why a recommendation was made.

-   3

    ### Maths and STEM tutoring

    Work through exercises in maths, physics or engineering, including problems given as a photo or diagram.

## Magistral Small 1.2: common questions

The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly.

Compliance depends on how you process data. On inference.eu, Magistral Small requests are processed in EU data centres, not used for training and covered by a data processing agreement under Article 28 GDPR with a published subprocessor list. That gives you a contractual basis to use it with personal data.

Magistral Small 1.2 has a 128K-token context window. Mistral notes that performance might degrade beyond about 40K tokens, so for very long documents you may get better results by splitting the input or choosing a long-context model such as Mistral Small 4.

Yes. Magistral Small is released under the Apache 2.0 licence, which allows commercial use, modification and redistribution without fees, user limits or revenue thresholds. You do not need a separate agreement with Mistral AI to build it into paid products.

Yes. Magistral writes its reasoning between thinking tags before the final answer. You can show it to users, log it for review or hide it and display only the answer. Remember that the reasoning counts as output tokens.

Yes, from version 1.2 onwards. Magistral Small 1.2 has a vision encoder, so it can reason about photos, diagrams and screenshots as well as text. Send images through the standard OpenAI \`image\_url\` content format.

Mistral Small 4 unifies reasoning, vision and coding in a larger mixture-of-experts model with a 256K context. It is the better default for most new work. Magistral Small remains a good choice when you want a compact, dense 24B model focused on reasoning.

Both are compact open reasoning models under Apache 2.0. gpt-oss-20b is a mixture-of-experts model with 3.6 billion active parameters and adjustable reasoning effort. Magistral Small is dense, comes from a European lab and reads images. Test both on your own tasks.

Mistral lists dozens of languages, including English, French, German, Greek, Italian, Polish, Portuguese, Romanian, Spanish, Serbian, Turkish, Ukrainian, Russian, Arabic, Chinese, Japanese and Korean. It can reason and answer in the language of the question.

Change your OpenAI client's base URL to the inference.eu API, use your inference.eu key and set the model to \`magistral-small\`. Allow a higher \`max\_tokens\` than for a non-reasoning model so the reasoning trace and answer both fit.

No. Requests to inference.eu, including any reasoning the model produces, are used only to produce your response. They are not used to train or fine-tune Magistral or any other model, and they are not shared with Mistral AI.

[Ask a question](https://inference.eu/contact/)

## Related

[All models](https://inference.eu/models/#directory)

-   [Mistral Small 4 119BRun Mistral Small 4 119B from French lab Mistral AI on inference.eu: reasoning, vision and code in one model, processed in EU data centres.Chat](https://inference.eu/models/mistral-small-4-119b/)
-   [Devstral Small 2Run Devstral Small 2, Mistral AI's open 24B coding agent model, on inference.eu with OpenAI and Anthropic APIs and processing in EU data centres.Code](https://inference.eu/models/devstral-small-2/)
-   [gpt-oss-20bRun OpenAI's compact open-weight gpt-oss-20b reasoning model on inference.eu: the OpenAI API format, processed in EU data centres, billed in euros.Reasoning](https://inference.eu/models/gpt-oss-20b/)
-   [Qwen3.8 27BRun Qwen3.8 27B on inference.eu: a dense open model with image and video input, adjustable reasoning and 256K context, OpenAI-compatible, processed in the EU.Reasoning](https://inference.eu/models/qwen3-8-27b/)

## Move your AI workloads to EU infrastructure

Tell us which models you use and roughly how much traffic you have. We reply within one working day with pricing and next steps.

-   Reply within one working day
-   Pricing for your volume
-   DPA ready to sign