Swiss AI
academicThe Swiss National AI Institute (SNAI) is a partnership between ETH Zurich's AI Center and EPFL's AI Center, launched in December 2023 with 10+ million GPU hours on the CSCS Alps supercomputer (270 petaflops, 10,752 H100 GPUs) and CHF 20M from the ETH Board. Over 70 AI professors and 800+ researchers across 10+ Swiss institutions. Strategic partner: Swisscom. Alliance with Stanford HAI.
The flagship Apertus (Sep 2025, 8B/70B) is trained on 15 trillion tokens across 1,811 languages (~40% non-English) using 4,096 GH200 GPUs on Alps. Uses xIELU activation, AdEMAMix optimizer, and the Goldfish objective to suppress verbatim memorization. Designed for EU AI Act compliance: trained exclusively on openly available data with retroactive robots.txt respect. 65K context. Apache 2.0. 101+ authors.
SNAI also produces domain-specific derivatives including Meditron (medical LLM, testing at Lausanne University Hospital) and multilingual research for underserved languages including Swiss German and Romansh.
People
- Antoine Bosselut Google ScholarOpenReview — Co-Lead, Swiss AI Initiative; Assistant Professor, EPFL
- Imanol Schlag Google ScholarOpenReview — Technical Lead, LLM Project; Research Scientist, ETH AI Center
- Martin Jaggi Google Scholar — Steering Committee; Associate Professor, EPFL