Switzerland has officially unveiled its own AI model called Apertus, which is positioned as an open alternative to OpenAI’s ChatGPT and Anthropic’s Claude. The name comes from the Latin word meaning “open”, and reflects the main idea of the project, reports The Verge.
The source code, training data, weights and a full description of the development process are already published on the HuggingFace platform.
The goal of Apertus developers is to create a new standard for reliable and globally relevant open models. AI was taught in more than 1,800 languages, and it was released in two versions — with 8 billion and 70 billion parameters. According to the characteristics, Apertus is comparable to the Llama 3 model from Meta.
In terms of scale, Apertus is one of the largest open-source models: it was trained on 15 trillion tokens in more than a thousand languages, including the Swiss dialect of German and Romansh. About 40% of the data is from non-English speaking sources, which makes the model more universal.
An important feature of the Swiss development was the strict adherence to European copyright legislation and the voluntary EU AI Code of Practice. In the US, technology corporations have repeatedly criticized these rules, believing that they hold back the development and implementation of artificial intelligence.
At the same time, the administration of US President Donald Trump is considering the possibility of introducing sanctions against European officials responsible for the implementation of the EU Digital Services Act (DSA). Washington sees the new regulation as imposing additional costs on American technology companies and restricting free speech. Last week, the US State Department discussed possible sanctions, but no final decision has yet been made.
