In an era of closed-source models, an open-source alternative made in Switzerland
As ChatGPT-style models proliferate and competition among tech giants intensifies, more and more voices are calling for greater transparency, control, and sovereignty in the field of generative artificial intelligence. While OpenAI, Anthropic, and Meta dominate the media landscape, their solutions remain, for the most part, centralized, opaque, or subject to commercial logic that is hardly compatible with the requirements of scientific or democratic transparency.
It is against this backdrop that Apertus has emerged—an unconventional and ambitious Swiss project that aims to offer an ethical, open-source, and sovereign alternative to ChatGPT. Built on a foundation of radical openness and local infrastructure, Apertus positions itself as a European response to the challenges of trust, accessibility, and inclusivity in generative AI.
A Swiss, open-source, community-driven project
Apertus is the result of a collaboration between Swiss academic institutions, open-source AI experts, and civil society organizations. The project is hosted locally in Switzerland on sovereign infrastructure, ensuring that data processing complies with European and Swiss standards.
Its uniqueness lies in its philosophy: fully open-source code, freely downloadable models, detailed documentation, and an explicit policy against the use of personal data for training. The stated goal is to enable any user, developer, researcher, or public decision-maker to examine, test, adapt, or integrate Apertus according to their own needs, without opacity or commercial dependence1.
By choosing a permissive license and a community-driven approach, the project’s founders aim to foster a local and transparent AI ecosystem that is open to European collaboration.
What can Apertus do? Current capabilities and use cases
Apertus functions as a chatbot similar to the general-purpose models available on the market. It can:
- answer open-ended questions,
- generate text, summarize documents, or rephrase content,
- assist with code writing or text analysis,
- adapt to specific corpora (legal, educational, medical fields, etc.).
From a technical standpoint, Apertus is based on a refined version of an existing open-source model, optimized for multilingualism (including French, German, and Italian), response stability, and the reduction of hallucinations. The system is enhanced by supervised fine-tuning, performed on verified corpora, with document filtering mechanisms and ethical safeguards2.
Compared to ChatGPT or Claude, Apertus performs less well on certain complex tasks, but stands out for its readability, transparency, and adaptability. It is already being used in several pilot projects in French-speaking Switzerland: educational support, chatbots for local governments, trials in public libraries, and writing assistance in public administrations.
Apertus vs. ChatGPT Comparison
| Criterion | Apertus | ChatGPT (GPT-4) |
| Origin | Switzerland (public / community) | United States (OpenAI, via Microsoft) |
| Access | Free and open source | Freemium (GPT-3.5 free, GPT-4 paid) |
| Bachelor's Degree | Open source (Apache/MIT) | Owner |
| Model transparency | Published code and weight | Not for public viewing |
| Main languages | Multilingual (French, German, Italian, etc.) | Multilingual (optimized for English) |
| Data hosting | In Switzerland (local self-government) | Azure Servers (US / Global) |
| Customization (fine-tuning) | Can be done by the user | Limited, via a paid API |
| Performance (complex tasks) | Average to good, depending on the corpus | Excellent (top-of-the-line model) |
| Ethics and Documentation | Transparent, with clearly stated limitations | Incomplete documentation, black box |
| Institutional use | Suitable for public and educational settings | Requires legal guidance |
Why Focus on Transparency? Apertus’ Ethical Choices
One of the project’s strengths lies in its commitment to documenting and explaining every stage of development. Unlike many closed-source models, Apertus provides:
- a template document detailing the parameters, the corpora used, and known limitations,
- access to the history of updates and patches,
- information on areas of uncertainty or potential biases in the responses.
The project is thus aligned with the principles of explainable and auditable AI, in line with European recommendations on algorithmic transparency3.
Apertus is not without its limitations. Performance may be lower on complex technical tasks, the model remains susceptible to certain linguistic biases, and the team cautions against unsupervised use in sensitive contexts. But here again, the project’s strategy is to highlight these limitations rather than hide them, and to encourage user feedback to improve the system.
Toward a more localized and democratic AI?
Apertus is part of a broader movement aimed at giving public institutions, researchers, and citizens real control over AI technologies. By offering an open, locally developed alternative, the project demonstrates that it is possible to create reliable, useful, and responsible chatbots without relying on centralized infrastructure or proprietary models.
This type of initiative could play a crucial role in European digital sovereignty, particularly by:
- supporting regional or underrepresented languages,
- enabling public institutions to develop their own in-house AI systems,
- promoting transparency and security in educational or institutional settings.
Apertus’s approach also demonstrates that making AI accessible to everyone requires readable code, clear intentions, and collective responsibility in the development of tools.
And tomorrow, ethical AI with a human face?
Apertus does not claim to immediately rival major proprietary models in terms of power or global reach. But it marks a significant shift in how we think about generative AI: no longer as a product, but as a technological commons.
In a landscape dominated by the race to build ever-larger models and a culture of opacity, projects like Apertus are bringing back essential values: transparency, accessibility, and citizen control. As the use of AI becomes more widespread, the question arises: what if the future of artificial intelligence lies in open, locally rooted, and collectively governed models?
Learn more
To broaden your understanding of the ethical issues surrounding AI, we recommend this article, which focuses on the fundamental principles: 7 Ethical Principles for Trustworthy Artificial Intelligence : it lays out the foundations of responsible AI, a theme closely linked to the philosophy of openness and transparency championed by the Apertus project
References
1. Apertus Project. (2025). Overview of the Apertus model.
https://apertus.ai/
2. Open Foundation Models. (2024). Comparative analysis of multilingual open LLMs.
https://ofm.eu.org/
3. European Commission. (2024). Transparency requirements for general-purpose AI models.
https://digital-strategy.ec.europa.eu/