Description
Apertus is a fully open multilingual foundation model for AI, developed by the Swiss AI Initiative in collaboration between EPFL, ETH Zurich and the Swiss National Supercomputing Centre. Its open-weights design is paired with open data, code, methods and alignment principles, all documented so that the work can be reproduced. The project calls itself open in the way that open source is open, and it describes responsible-scale techniques for respecting opt-outs, removing personal information and preventing memorisation. The model comes in 8B and 70B parameter sizes, and the site says it is competitive with top open models at those scales. It is multilingual from the start and was trained on data in more than 1000 languages. The latest release, Apertus 1.5, adds multimodal input, reasoning, longer context and better instruction following. There is an inference ecosystem for getting started and a technical report presented at ACL 2026. The name comes from the Latin word for open. It suits researchers, developers and public institutions that want a transparent foundation model.…
Highlights
- Open weights, data, code and methods
- 8B and 70B parameter models
- Multilingual, trained on over 1000 languages
- Apertus 1.5 with multimodal input and reasoning
- Technical report presented at ACL 2026
Ratings & Reviews
No ratings yet
Be the first to rate Apertus.
People also visit
Apertus FAQ
What is Apertus?
A fully open multilingual foundation model for AI, built by the Swiss AI Initiative with EPFL, ETH Zurich and CSCS.
What is open about it?
The site says the training data, code, weights, methods and alignment principles are all documented and reproducible.
What sizes are available?
The model comes in 8B and 70B parameter versions, according to the site.
What is new in Apertus 1.5?
Multimodal input, reasoning, longer context and better instruction following.
What are the best Apertus alternatives?
Popular alternatives to Apertus include Kimi K2.5, NExT-GPT and OmAgent. See all Apertus alternatives

