Apertus 8B Instruct
Swiss AI Initiative · 2025-09-02 · 8.1B parameters
Apertus 8B Instruct is the instruction-tuned 8-billion-parameter variant of the Apertus model family, developed by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS) as part of the Swiss AI Initiative, released in September 2025. Its base model was pretrained from scratch on 15 trillion tokens across more than 1,000 languages using entirely open and compliant training data, with a novel xIELU activation function and AdEMAMix optimizer. It targets multilingual dialogue and instruction following with a focus on auditability and European data sovereignty.