A university contribution to open model development
EPFL traces its roots to the École Spéciale de Lausanne, founded in 1853, and adopted its federal polytechnic name in 1969. Its role in Apertus is that of a model co-developer, rather than merely the affiliation of a scientist quoted in a launch announcement. [1][2]
Joint development and public research resources
The September 2025 Apertus release joined ETH Zurich, EPFL and CSCS in developing language models trained on the Alps supercomputer. The release made 8B and 70B model weights available with code and training documentation, giving researchers material to inspect and adapt outside a hosted chat service. [2]
The 8B instruction-tuned model card identifies its weights as Apache 2.0. That model license does not establish a uniform license for every item in the underlying training corpus. The release is evidence of a public model artifact, while its authors' capability claims remain their own evaluations. [3][2]
In July 2026, the partners announced Apertus 1.5, adding image and audio understanding to the project’s text-model work. CSCS also introduced an inference service for its users and the Swiss academic community. This institutional access route is distinct from making model files downloadable. [4]
Sources
- Overview
EPFL · retrieved Oct 23, 2026
- Apertus: A fully open, transparent, multilingual language model
Swiss National Supercomputing Centre · Sep 2, 2025
- swiss-ai/Apertus-8B-Instruct-2509
Swiss AI Initiative · retrieved Oct 23, 2026
- Apertus 1.5: Building the next generation of open AI infrastructure
Swiss National Supercomputing Centre · Jul 24, 2026