Switzerland has launched Apertus, a fully open-source multilingual AI model trained on 15 trillion tokens across 1,000+ languages. Built by EPFL, ETH Zurich, and CSCS with Swisscom’s support, Apertus sets new standards for transparency, inclusivity, and digital sovereignty in the AI world.
A consortium of Swiss institutions unveiled Apertus, a groundbreaking open-source artificial intelligence (AI) model poised to redefine the landscape of generative AI. Developed collaboratively by École Polytechnique Fédérale de Lausanne (EPFL), ETH Zurich, and the Swiss National Supercomputing Centre (CSCS), Apertus—Latin for “open”—embodies a commitment to transparency, inclusivity, and digital sovereignty. This initiative, supported by Swisscom, marks a significant milestone in creating a trustworthy AI framework accessible to researchers, developers, and organizations worldwide.
A Foundation for Open AI
Apertus distinguishes itself in the crowded AI ecosystem by offering complete transparency in its design and development. Unlike many large language models (LLMs) that obscure critical details about their architecture or training processes, Apertus lays bare its model weights, training data, and documentation. Available in two versions—an 8-billion-parameter model and a more robust 70-billion-parameter version—it is released under a permissive open-source license, enabling its use in research, education, and commercial applications.
This openness is not merely symbolic. “With this release, we aim to provide a blueprint for how a trustworthy, sovereign, and inclusive AI model can be developed,” said Martin Jaggi, Professor of Machine Learning at EPFL and a member of the Swiss AI Initiative’s Steering Committee. The model’s architecture is designed to foster innovation while adhering to strict ethical and legal standards, including Swiss data protection laws and the transparency mandates of the EU AI Act. By making every component accessible, Apertus empowers developers to create applications ranging from chatbots to translation tools and education-focused platforms.
The model is hosted on Hugging Face, a widely used platform for AI model distribution, and can also be accessed through Swisscom’s sovereign AI infrastructure. This dual-access approach ensures that both individual developers and large organizations can leverage Apertus for their needs. For those requiring computational resources, the Public AI Inference Utility provides international access, while Swisscom’s platform caters to business customers seeking secure, localized solutions.
Multilingual Excellence
One of Apertus’s standout features is its multilingual capability, trained on a staggering 15 trillion tokens across more than 1,000 languages. Approximately 40% of its training data comprises non-English languages, including underrepresented ones like Swiss German and Romansh. This focus on linguistic diversity sets Apertus apart from many LLMs, which often prioritize English and a handful of major languages. By incorporating such a broad linguistic spectrum, Apertus ensures that AI applications can serve diverse communities, from local Swiss populations to global users.
“Apertus is built for the public good,” said Imanol Schlag, the project’s technical lead and a Research Scientist at ETH Zurich. “It stands among the few fully open LLMs at this scale and is the first of its kind to embody multilingualism, transparency, and compliance as foundational design principles.” This emphasis on inclusivity aligns with Switzerland’s broader mission to create AI that serves as public infrastructure, akin to essential utilities like highways or electricity.
The training process itself reflects a meticulous approach to data integrity. The dataset, drawn exclusively from publicly available sources, was rigorously filtered to exclude personal data and respect website opt-out requests. Ethical guidelines further ensured that undesirable or harmful content was removed before training, aligning with Switzerland’s reputation for precision and accountability.
Infrastructure for Innovation
Apertus is not a one-off product but a long-term commitment to building AI infrastructure. Thomas Schulthess, Director of CSCS and Professor at ETH Zurich, described it as “a driver of innovation and a means of strengthening AI expertise in research, society, and industry.” Unlike traditional technology transfers from research to market, Apertus is designed as a sustainable resource, with regular updates planned by a dedicated team of engineers and researchers from CSCS, ETH Zurich, and EPFL. The model’s training was carried out on Switzerland’s Alps supercomputer at CSCS, a state-of-the-art facility that underscores the project’s technical rigor and computational prowess.
Swisscom’s role as a strategic partner further strengthens this vision. By integrating Apertus into its sovereign AI platform, the telecommunications giant is fostering a secure and responsible AI ecosystem. “This underscores our commitment to shaping a secure and responsible AI ecosystem that serves the public interest and strengthens Switzerland’s digital sovereignty,” said Daniel Dobos, Research Director at Swisscom. The company’s infrastructure ensures that businesses can deploy Apertus seamlessly, while developers can experiment with the model during events like the Swiss {ai} Weeks, which run through October 5, 2025.
For developers, accessing Apertus is straightforward but requires adequate computational resources. The 8-billion-parameter model is suitable for smaller-scale applications, while the 70-billion-parameter version caters to more complex tasks requiring significant processing power. During the Swiss {ai} Weeks, hackathon participants can test Apertus through a Swisscom-hosted interface, democratizing access to this powerful tool. International users, meanwhile, can tap into the Public AI Inference Utility, ensuring global reach.
Transparency and Compliance
Transparency is at the heart of Apertus’s design. Beyond making its architecture and training data publicly available, the project adheres to stringent ethical and legal standards. The training process complies with Swiss data protection regulations, Swiss copyright law, and the EU AI Act’s transparency requirements. Intermediate checkpoints and model weights are also accessible, allowing researchers to study and refine the model’s development process.
This commitment to compliance is particularly significant in an era of increasing scrutiny over AI ethics. By limiting the dataset to publicly available information and applying rigorous filtering, Apertus sets a high standard for responsible AI development. Joshua Tan, Lead Maintainer of the Public AI Inference Utility, emphasized this point: “Currently, Apertus is the leading public AI model: a model built by public institutions, for the public interest. It is our best proof yet that AI can be a form of public infrastructure.”
A Vision for the Future
Apertus is not a static achievement but the foundation of an evolving ecosystem. “The release of Apertus is not a final step, rather it’s the beginning of a journey,” said Antoine Bosselut, Professor at EPFL and Co-Lead of the Swiss AI Initiative. Future updates will expand the model family, improve computational efficiency, and develop domain-specific tools for fields like law, health, climate, and education. These advancements will maintain the project’s commitment to transparency and ethical standards while addressing real-world challenges.
For instance, domain-specific adaptations could enable Apertus to power legal research platforms, assist in climate modeling, or support personalized education tools. By fostering such applications, Switzerland is positioning itself as a global leader in responsible AI innovation. The project’s emphasis on openness also invites collaboration, encouraging researchers and developers worldwide to contribute to its evolution.
Global Implications
Apertus arrives at a critical juncture in the global AI landscape, where proprietary models from companies like OpenAI, Anthropic, and Google DeepMind dominate with restricted access and opaque development processes. By contrast, Apertus’s fully open approach challenges this paradigm, offering a powerful, transparent alternative that prioritizes public interest over corporate control. With 15 trillion tokens and support for over 1,000 languages, Apertus rivals some of the largest LLMs in scale while surpassing many in transparency and inclusivity.
The project also aligns with Switzerland’s broader digital strategy. By prioritizing digital sovereignty, Apertus ensures that critical AI infrastructure remains under public control, reducing reliance on foreign technology providers. This approach resonates globally, as nations increasingly seek to balance technological advancement with ethical considerations and data sovereignty.
For developers, Apertus offers a versatile platform for innovation. Its open-source license allows for unrestricted experimentation, whether for creating multilingual chatbots, translation tools, or educational applications. The availability of two model sizes ensures flexibility, catering to both resource-constrained environments and high-performance computing needs.
Apertus represents a bold step forward in the quest for trustworthy, inclusive, and transparent AI. By combining cutting-edge technology with a commitment to openness, Switzerland has created a model that serves as both a tool and a template for responsible AI development. As the project evolves, it promises to drive innovation across industries, empower diverse communities, and redefine AI as a public good. For researchers, developers, and organizations worldwide, Apertus is not just an AI model—it’s an invitation to build a better, more inclusive future.