Apertus: Open Foundation Model for Sovereign AI
Original: Apertus – Open Foundation Model for Sovereign AI
Why This Matters
Demonstrates European approach to sovereign AI with full transparency and regulatory compliance, competing with proprietary models.
Swiss AI Initiative launches Apertus, fully open foundation model developed by EPFL, ETH Zurich, and CSCS with open weights, data, and code. Available in 8B and 70B parameters, multilingual across 1000+ languages, EU AI Act compliant.
Apertus is a fully open foundation model for sovereign AI developed through a collaborative effort between Swiss institutions including EPFL, ETH Zurich, and CSCS, under the Swiss AI Initiative. The model features complete transparency with open weights, open training data, open code, and documented methods and alignment principles, making it reproducible and auditable. The project aims to provide a global foundation comparable to proprietary alternatives. Apertus is built to comply with EU AI Act requirements, including opt-out functionality, personally identifiable information removal, and memorization prevention. Performance metrics show competitiveness with top open models at equivalent scales of 8B and 70B parameters. The model supports multilingual capabilities across 1000+ languages from inception. Recent developments include the June 2026 release of Apertus Mini, a set of 16 small language models demonstrating distillation and quantization techniques. A technical paper on Apertus was accepted at ACL 2026, a leading conference for AI and natural language processing. Swisscom is identified as a strategic partner of the Swiss AI Initiative.