Apertus is the open source 8b and 70b LLM from swiss-ai. They've published both the base and the instruct sft models. Very cool that projects like this exist.
I haven't tried it for anything myself yet. The paper provides several benchmarks. The emphasis during training was on multi-language support (over 1800 languages are represented in its pre-training data, which is 40% non-English) and non-copyrighted training data... and the benchmarks seem to suffer for it.
Yes it’s not bad, although it’s not meant to be a chatbot, post training is limited, so it won’t feel as smooth as TOTL of course. The number of supported languages is mind boggling.
Focus was on open data, languages and auditability.
Their loss function is fancy, not sure about the effects
Not really. It's because the target audience is more academic/scientific rather than the Swiss population at large. In the latter case, it would be in the local languages. The law is relatively clear for this. English is not accepted in Switzerland as a replacement language for the "local" ones, although many people can speak or at least understand some English.
What if I told you there’s this thing in 2026 called an LLM that can translate between any two languages with high fidelity for free, and you just clicked a single button in your browser to use it
Not quite: it's a collab between both ETHZ (Zürich, German speaking) and EPFL (Lausanne, French speaking). According to the website, the actual hardware is distributed all over the country (including in the Italian part).
Apertus is the open source 8b and 70b LLM from swiss-ai. They've published both the base and the instruct sft models. Very cool that projects like this exist.
https://apertvs.ai/pages/documentation/
Tech report:
https://arxiv.org/pdf/2509.14233
Is it any good?
I haven't tried it for anything myself yet. The paper provides several benchmarks. The emphasis during training was on multi-language support (over 1800 languages are represented in its pre-training data, which is 40% non-English) and non-copyrighted training data... and the benchmarks seem to suffer for it.
https://arxiv.org/abs/2509.14233
it's quite bad tbh. i've tried it for some time and i expected much more...
Yes it’s not bad, although it’s not meant to be a chatbot, post training is limited, so it won’t feel as smooth as TOTL of course. The number of supported languages is mind boggling.
Focus was on open data, languages and auditability.
Their loss function is fancy, not sure about the effects
2023, but deadlines less than a month ago? Seems to be been updated continiously so (2023) doesn't really fit here.
I propose every Linux post should be tagged (1991) from now on
(2023) Little said at the time (4 points, 1 comment) https://news.ycombinator.com/item?id=38529956
Related 2023 discussion (22 comments): <https://news.ycombinator.com/item?id=38523736>
My standard question about Swiss engineering, “how many jewels?”, is failing me.
Has anything noteworthy come from this initiative? I have not heard of anything yet.
Why is this not written in German, I'm afraid to ask?
Why is it not written in French? Or Italian? Or Romansh? Because Switzerland has four official languages and English makes it easier for everyone
Not really. It's because the target audience is more academic/scientific rather than the Swiss population at large. In the latter case, it would be in the local languages. The law is relatively clear for this. English is not accepted in Switzerland as a replacement language for the "local" ones, although many people can speak or at least understand some English.
heavy sigh I'm Swiss. I know. What I meant to say is that German is not the default language in Switzerland.
Most researchers in Switzerland are non-Swiss, and many institutes have English as language of business
Staff nationality of Swiss higher education institutions:
- Universities: 55% Swiss, 45% foreign - Universities of applied sciences: 75% Swiss, 25% foreign - Universities of teacher education: 87% Swiss, 13% foreign - Professors: 49% Swiss, 51% foreign - PhDs/scientific collaborators: 30% Swiss, 70% foreign - Professors of ETH Zurich: 31% Swiss, 69% foreign
english is the lingua franca
Why it has to be german?
What if I told you there’s this thing in 2026 called an LLM that can translate between any two languages with high fidelity for free, and you just clicked a single button in your browser to use it
It's a university in a French speaking region for one.
Not quite: it's a collab between both ETHZ (Zürich, German speaking) and EPFL (Lausanne, French speaking). According to the website, the actual hardware is distributed all over the country (including in the Italian part).
because the brits won the language wars.
And the other wars ;)
Because german is hard.