Swiss AI Initiative Releases Apertus, a Fully Open Foundation Model for Sovereign AI
Tags AI · OSS · Infrastructure
The Swiss AI Initiative — a collaboration between EPFL, ETH Zurich, and CSCS — launched Apertus, a fully open foundation model available in 8B and 70B parameter versions. The model is trained on over 1,000 languages and built to comply with the EU AI Act, respecting opt-outs, removing PII, and preventing memorization. Unlike most frontier models, Apertus makes training data, code, weights, methods, and alignment principles all public and reproducible. Swisscom is a strategic partner. The model launched on June 19, 2026 and quickly gained 392 points on Hacker News. This is the first major European effort to build a sovereign AI foundation model that can compete with US and Chinese alternatives on capability while meeting stricter regulatory requirements.
Technical significance
Apertus represents Europe's most credible attempt at a sovereign AI foundation model. If benchmark results confirm competitiveness with Llama 3 and Mistral at equivalent scales, it gives European enterprises and governments a non-US, non-Chinese alternative that is natively EU AI Act compliant. The 1,000+ language training is a differentiator for multilingual European deployments.