Apertus, open to all

Apertus is a fully open large language model developed in Switzerland

downloads on Hugging Face

5M+

supported languages

1,800+

tokens of training data

15 trillion

parameters in the largest model

70B

From Swiss research

Developed by the Swiss AI Initiative around EPFL, ETH Zurich and the CSCS supercomputing centre

Fully open

Model weights, training recipe and data basis are openly accessible and auditable

Multilingual and built for Switzerland

Apertus was trained on more than 1,800 languages, around ten times more than comparable language models. This is less about translation quality than about multicultural representation: many cultures are reflected in the training data, including all of Switzerland's national languages.

Trained in Switzerland

Trained on the Alps supercomputer at CSCS in Lugano

Use Apertus in the chat →→ More about Apertus→