Evidence-backed European AI register
Swiss AI Initiative (EPFL + ETH Zurich + CSCS)
Reviewed identity, roles, model listings, pricing and evidence for Swiss AI Initiative (EPFL + ETH Zurich + CSCS).
Dataset compiled 2026-08-28 · evidence over inference
Reviewed identity
- Headquarters scope
- EU body or consortium
- Headquarters country
- unknown
- Headquarters city
- unknown
- Legal entity
- university consortium — EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS), with 10+ Swiss academic institutions
- Ownership
- public research
- Founded
- 2023
- Confidence
- high
- Last verified
- 2026-08-22
Headquarters evidence
The initiative is the largest open science/open source effort for AI foundation models worldwide, and the first initiative of the Swiss National AI Institute, a partnership between the ETH AI Center and the EPFL AI Center.
The tracked subject is an institutional partnership/initiative, not a separately headquartered company.
Evidence boundaries
Organisation listings identify a reviewed canonical record. Headquarters is independent of incorporation, registered offices, ownership, hosting and runtime. Model listings show an association in the source register; they do not by themselves establish model creation, current API availability, hosting location or infrastructure ownership.
Model listings
- Type
- text
- Parameters
- 70B (71B on HF)
- Context
- 65,536 tokens
- Languages
- 1,000+ languages; 15T training tokens with ~40% non-English, including Swiss German and Romansh
- Licence
- Apache-2.0
- Release
- 2025-09-02
- Type
- text
- Parameters
- 8B
- Context
- 65,536 tokens
- Languages
- 1,000+ languages
- Licence
- Apache-2.0
- Release
- 2025-09-02
- Type
- multimodal
- Parameters
- 72B
- Context
- unknown
- Languages
- globally multilingual, multimodal input with reasoning support
- Licence
- Apache-2.0
- Release
- 2026
- Type
- multimodal
- Parameters
- 9B
- Context
- unknown
- Languages
- multilingual multimodal
- Licence
- Apache-2.0
- Release
- 2026
- Type
- text
- Parameters
- 4B / 2B / 0.4B
- Context
- unknown
- Languages
- multilingual
- Licence
- Apache-2.0
- Release
- 2026
- wavtokenizer-large-unify-40token
- Type
- asr
- Parameters
- 76.3M
- Context
- unknown
- Languages
- n/a — audio tokeniser / feature extraction
- Licence
- Apache-2.0
- Release
- unknown
Profile note
The most transparent large open model in Europe and Switzerland's national AI project: Apertus publishes not only weights but training data composition, recipes and intermediate checkpoints under Apache-2.0, trained entirely on the publicly owned CSCS Alps supercomputer across 1,000+ languages including Swiss German and Romansh. It has no commercial arm and no pricing of its own — monetisation happens downstream at Swisscom, Infomaniak and the Public AI Inference Utility. The single most important reference point for what 'sovereign European AI' can mean in practice.
Record sources
- https://ethz.ch/en/news-and-events/eth-news/news/2025/09/press-release-apertus-a-fully-open-transparent-multi…
- https://huggingface.co/swiss-ai
- https://huggingface.co/swiss-ai/Apertus-70B-2509
- https://www.swisscom.ch/en/about/news/2025/09/02-apertus.html
- https://publicai.co/stories/apertus
- https://en.wikipedia.org/wiki/Apertus_(LLM)
- https://ethz.ch/