Switzerland has officially launched Apertus, a large, open-source language model created by government institutions. It is effectively the country’s first national LLM model, positioned as a public infrastructure accessible to everyone, from researchers to businesses.
Switzerland has officially launched Apertus, a large, open-source language model created by government institutions. It is effectively the country’s first national LLM model, positioned as a public infrastructure accessible to everyone, from researchers to businesses.
According to Engadget, the development was led by the Federal Institute of Technology Lausanne (EPFL), ETH Zurich and the Swiss National Supercomputing Center (CSCS). The model was released in two versions — with 8 billion and 70 billion parameters. Access is available through Swisscom or on the Hugging Face platform.
Apertus was trained on 15 trillion tokens in over a thousand languages, of which about 40% are non-English texts. The dataset also includes Swiss languages, including German dialects and Romansh. The team emphasizes that only open sources were used for training, and the crawlers respected the sites’ settings to prohibit indexing. This is an important difference from many commercial models, which are often accused of copyright infringement and unfair data collection.
Apertus is characterized by complete transparency. The developers have opened up not only the model, but also the training code, documentation, and a list of sources used. This approach makes it convenient for companies that must comply with European data protection and copyright laws. Banks are already showing interest in the project, as a local solution can comply with strict financial confidentiality rules.
The launch of Apertus illustrates a new approach by states to AI: instead of relying on private corporations, countries are seeking to create their own open models as strategic infrastructure. This is particularly important for Europe, as it concerns data control and regulatory compliance. If the Swiss initiative is successful, other EU states may follow suit and start working on their own LLM models.
As a reminder, dev.ua previously also wrote about how the Ministry of Digital Affairs, together with Kyivstar, gathered leaders of four key areas — from technical to cultural and historical. The team includes well-known scientists, lawyers, and managers who will determine the standards and vision of the Ukrainian analogue of ChatGPT.
