Apertus Swiss AI Offers Data Protection and Open Source Code

**Apertus** is a Swiss open-source AI model designed for privacy and transparency, offering a sovereign alternative to large commercial models. It is already in use by public administration and available for developers.

## Swiss AI with a Public Face

A new player in the field of artificial intelligence, **Apertus**, emerges from **Switzerland**. It is not another commercial giant accustomed to its "black boxes." This language model (LLM) arrives with a radically different approach: public, academic, and above all, open source. The idea is clear: to offer a sovereign and transparent alternative to names like **ChatGPT**, **Gemini**, or **Claude**. Behind this is the **Swiss AI Initiative**, which has driven this project with a focus on auditability and data control.

## Power and Linguistic Diversity

**Apertus** is powerful. It comes in two versions: a lightweight version with **8 billion parameters** and a more potent one with **70 billion**. Its training utilized **15 trillion tokens**, nothing less, leveraging the power of the **Alps** supercomputer at the Swiss National Centre for Supercomputing. But what truly stands out is its multilingual vocation. It has been trained on over **1,000 languages**, and nearly **40%** of its training data is not in English. It natively includes languages often ignored by big tech companies, such as **Romansh** and **Swiss German**.

## Transparency and Swiss Values

The fundamental difference from proprietary models is the total openness of **Apertus**. Its architecture, weights, code, and documentation are public and downloadable. This means companies and institutions can run it on their own servers, maintaining absolute control over their data. Furthermore, it has undergone rigorous filtering to comply with copyright and data protection laws, aligning with the **European Union AI Act**. Principles such as neutrality, diversity, and public service vocation have been injected into it—like Swiss DNA applied to AI.

## Applications and Open Future

The project is already making an impact in real environments. The **Canton of Ticino** officially uses it for document translation. It is available free on the **Hugging Face** platform, where it has already accumulated over **4 million downloads**, and through services like **Swisscom** and **Infomaniak**. The future is ambitious: scientists and developers are already working on adapting it to highly sensitive sectors such as health, climate, law, and education.