---
title: "Devstral Small 2 API hosted in the EU"
description: "Run 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."
canonical: "https://inference.eu/models/devstral-small-2/"
language: "en"
---

Code · Mistral AI

# Run Devstral Small 2 in the EU for coding agents

Devstral Small 2 is Mistral AI's open model for software engineering agents: 24 billion parameters, a 256K context and image input, released under Apache 2.0. On inference.eu you call it with the OpenAI or Anthropic request format, so coding tools can use it with your source code 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 Devstral Small 2 is

Devstral Small 2, published as Devstral-Small-2-24B-Instruct-2512 in December 2025, is an agentic coding model from Mistral AI in Paris. It is built to use tools: exploring a repository, reading and editing several files, running commands and reacting to the output, which is how modern coding agents work.

It has 24 billion parameters, a context window of 256K tokens and, new in version 2, a vision encoder so it can read screenshots and diagrams. Mistral released it alongside the larger Devstral 2 and its own terminal tool, Mistral Vibe. Devstral Small 2 is licensed under Apache 2.0; the larger Devstral 2 uses a different licence.

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

Devstral Small 2 is strong for its size at agent-style software tasks: fixing bugs across files, writing tests, refactoring and following a codebase's conventions. The 256K context lets an agent keep large parts of a repository and long tool histories in view. Its compact size makes it a good fit for teams that want a coding model they could also run on their own hardware later.

Larger models such as GLM-5.3, Kimi K3 or DeepSeek V4 go further on the hardest long-horizon tasks. Devstral is tuned for code, not for general writing or broad world knowledge. It has no separate thinking mode, so for algorithmic puzzles a reasoning model can be the better partner.

## Using Devstral Small 2 in coding tools

inference.eu offers both the OpenAI and the Anthropic request formats. Tools that accept an OpenAI-compatible endpoint, such as Cline, Kilo Code, OpenHands, SWE-agent or Mistral Vibe, work by setting the base URL to the inference.eu API, adding your inference.eu key and choosing the model `devstral-small-2`.

Tools that speak the Anthropic Messages format, such as Claude Code, can point at the same model through our Anthropic-compatible endpoint. Tool calls and streaming work as usual. Send screenshots of a failing UI as `image_url` parts. Keep agent loops in check with step limits, since long sessions consume context quickly.

## Your source code stays in the EU

Source code is often a company's most sensitive asset. With Devstral Small 2 on inference.eu, the code your agents read, the diffs they write and the commands they run are processed in data centres in the European Union, and none of it is used to train any model.

Every customer gets a data processing agreement under Article 28 of the GDPR and our subprocessor list, which your security team can review before rollout. Billing is in euros. The Apache 2.0 licence allows commercial use of the model without limits, so you can use it for client projects and in-house products alike.

## Devstral Small 2

Maker

Mistral AI

Made in

France

Type

Code

Parameters

24B

Context window

262,144 tokens

Licence

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

Released

December 2025

Model ID

devstral-small-2

Processing

EU data centres

Training on your data

Never

Source

[huggingface.co](https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512)

devstral-small-2 · 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="devstral-small-2",
    messages=[{"role": "user", "content": "Summarise this contract in five bullet points."}],
)
print(response.choices[0].message.content)
```

## What teams use it for

-   1

    ### Coding agents in the terminal

    Let an agent explore a repository, fix a failing test and propose a patch, with every request processed in the EU.

-   2

    ### Code review support

    Summarise pull requests, flag risky changes and suggest missing tests before a human reviewer looks at the diff.

-   3

    ### Legacy code migration

    Upgrade dependencies, port code between frameworks and update call sites across many files in one agent run.

## Devstral Small 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, Devstral Small 2 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 covers source code containing personal data, such as test fixtures or logs.

Yes. inference.eu offers an Anthropic-compatible endpoint, so tools built for the Anthropic Messages format, including Claude Code, can use Devstral Small 2 by changing the base URL, API key and model name. Mistral lists Claude Code among the scaffolds Devstral works with.

Devstral Small 2 has a 256K-token context window. That lets an agent keep many files, long tool outputs and a lengthy history of steps in view. Very long agent sessions still fill it up, so most tools compact or trim history over time.

Yes. Devstral Small 2 is released under the Apache 2.0 licence, which allows commercial and non-commercial use and modification. The larger Devstral 2 123B model is under a different, more restrictive licence, so check that one separately.

Devstral 2 has 123 billion parameters and scores higher on software engineering benchmarks. Devstral Small 2 has 24 billion, is much lighter and uses the permissive Apache 2.0 licence. For most day-to-day agent work, the small model is a sensible starting point.

Mistral lists Mistral Vibe, Cline, Kilo Code, Claude Code, OpenHands and SWE-agent. Any tool that lets you set an OpenAI-compatible or Anthropic-compatible base URL and model name can use it through inference.eu.

Yes. Version 2 added vision, so the model can look at screenshots of a broken layout, error dialogs or architecture diagrams alongside code. Send images as \`image\_url\` content parts in the OpenAI format.

Both are compact models aimed at agentic coding with 256K context and image input. Qwen3.6 27B adds a thinking mode; Devstral Small 2 comes from a European lab and is tuned specifically for tool-using software agents. Try both inside your own agent setup.

No. Code and prompts sent to inference.eu are used only to produce your response. They are never used to train or fine-tune Devstral or any other model, and they are not shared with Mistral AI.

In your tool's settings, set the base URL to the inference.eu API, enter your inference.eu key and choose \`devstral-small-2\` as the model. Tool calling uses the standard OpenAI format, so agent behaviour carries over.

[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/)
-   [Qwen3.6 27BRun Qwen3.6 27B on inference.eu: a dense open model for agentic coding with image input and 256K context, OpenAI-compatible and processed in the EU.Code](https://inference.eu/models/qwen3-6-27b/)
-   [Magistral Small 1.2Run 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.Reasoning](https://inference.eu/models/magistral-small/)
-   [GLM-5.3Run Z.ai's GLM-5.3, an open-weight model built for complex coding and long agent tasks, on inference.eu with OpenAI and Anthropic APIs, processed in the EU.Reasoning](https://inference.eu/models/glm-5-3/)

## 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