From Switzerland: Apertus, a New Open-Source Language Model
On September 2, 2025, the Swiss AI Initiative – a collaboration between EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS) – released Apertus, Switzerland’s first large-scale open-source language model. The project aims to combine transparency, multilingual capabilities, and accessibility, positioning Switzerland as a contributor to global artificial intelligence research.
What Is Apertus?
Apertus is a large language model (LLM) released in two versions: 8B and 70B parameters, both of which are licensed under the Apache 2.0 open-source license. Unlike many proprietary models, Apertus provides full access to its training architecture, datasets, model weights, and intermediate checkpoints. The source code, deployment guides, and logs are also openly available, making the model entirely transparent and reproducible.
According to the project team, this design enables developers, researchers, and institutions to examine and modify every part of the model. The initiative emphasizes openness as a guiding principle, reflected in the choice of name – Apertus, derived from Latin, meaning “open.”
Multilingual Training
One of its defining features is its broad linguistic coverage. The model was trained on 15 trillion tokens across more than 1,500 languages, with 40% of the data in non-English languages. This includes Switzerland’s own linguistic heritage, such as Romansh and Swiss German, as well as other languages often underrepresented in mainstream AI models.
The project highlights this approach as an effort to serve a wider global community, moving away from the typical English-first training focus seen in other models.
Research and Applications
The 8B version of Apertus is designed for local use, fine-tuning, and smaller-scale research. The 70B version is intended for large-scale deployments and more resource-intensive applications. Both models can be applied in areas such as:
- Translation and multilingual tools
- Educational platforms and tutoring systems
- Summarization and information retrieval
- Conversational systems and chatbots
Because Apertus is licensed under Apache 2.0, it can be freely used for both research and commercial projects.
Global Support and Deployment
The international computing community supported the public release. Public AI, the official deployer of Apertus, allocated over 115,000 GPU-hours across 20 clusters in five countries for the launch month. Infrastructure contributions were made by partners including AWS, Exoscale, AI Singapore, Cudo Compute, CSCS, and NCI Australia.
Users can access Apertus through the Public AI Inference Utility, on Hugging Face, or directly via Swisscom in Switzerland.
The Swiss AI Initiative considers Apertus only the beginning. Future plans include domain-specific adaptations in fields such as healthcare, climate, law, and education, while maintaining its founding principles of openness and transparency.
In a statement, project co-lead Antoine Bosselut described Apertus as a “long-term commitment to sovereign, open AI foundations that serve the public good, worldwide.”
Looking Ahead
With Apertus, Switzerland has entered the global AI stage with a model that reflects its national values of transparency, multilingualism, and public service. By making the model fully open and accessible, the Swiss AI Initiative seeks to provide researchers, educators, and businesses with a foundation that is both practical and principled.
