The Swiss AI Initiative, a collaboration between EPFL, ETH Zurich, and CSCS, has launched Apertus, a fully open foundation model designed for sovereign AI. Apertus features open weights, data, code, and alignment principles, aiming to provide a transparent and reproducible AI platform. The model is available now and supports over 1,000 languages, making it one of the most multilingual open models to date, according to apertvs.ai.
Apertus was developed with compliance to the EU AI Act in mind, incorporating mechanisms to respect opt-outs, remove personally identifiable information, and prevent data memorization. The model is competitive with other open models at scales of 8 billion and 70 billion parameters. Swisscom acts as a strategic partner, supporting the Swiss AI Initiative’s goal of creating a sovereign and open AI ecosystem, as detailed on apertvs.ai.
The launch of Apertus addresses growing concerns around AI transparency, data privacy, and regulatory compliance in Europe. By providing an open-source alternative that meets stringent EU requirements, Apertus positions itself alongside other open models like Meta’s LLaMA and EleutherAI’s GPT-NeoX, but with a stronger emphasis on sovereignty and ethical AI use. This initiative reflects increasing governmental and institutional interest in AI models that can be audited and controlled locally.
Apertus is accessible to developers and researchers through the Swiss AI Initiative’s platform, with full documentation and technical details available online. The initiative also offers a newsletter to keep the community updated on releases and research, reinforcing its commitment to open science and collaborative AI development.