#apertus — Public Fediverse posts

#Apertus was also mentioned earlier this year in this WIRED global story https://www.wired.com/story/europe-race-us-deepseek-sovereign-ai/

#Apertus featured in #WIRED Italy https://www.wired.it/article/intelligenza-artificiale-svizzera-apertus-modelli/

#Apertus featured in #WIRED Italy https://www.wired.it/article/intelligenza-artificiale-svizzera-apertus-modelli/

#Apertus featured in #WIRED Italy https://www.wired.it/article/intelligenza-artificiale-svizzera-apertus-modelli/

#Apertus featured in #WIRED Italy https://www.wired.it/article/intelligenza-artificiale-svizzera-apertus-modelli/

#Apertus featured in #WIRED Italy https://www.wired.it/article/intelligenza-artificiale-svizzera-apertus-modelli/

Come join the #Apertus LLM team as an AI research engineer!

If you have experience with software, data, and ML engineering, a passion for #FOSS and interesting in post-training of large models (#SFT, #RL, rewards design, ...), you could be a great fit for the recently opened roles, all in Lausanne, Switzerland!

https://careers.epfl.ch/job/Lausanne-AI-Research-Engineers-Apertus-Initiative/1164610655/

#Fedihire #Job #Swtizerland #FOSS #ML #AI

so my #Nextcloud classifier only needs some last minir tweaks.

#Apertus 8B was too much. responses took over an hour. with #Mistral 3B its 15-20mins. which is fine since I only have a few documents max per day. I guess if I ever need to process more, I could just point the classifier at ollama on my laptop.

also hooked up my #ix500 scanner to the #Pi4 and its scanning multi page documents.

I have a daily notification email with links to a manual review UI to tweak any issues manually.

Working on my #Nextcloud classifier project. While #Apertus 8B Q4_K_S did fit into the memory of my #pi4 8GB even overclocked to 2000Mhz it wasn't able to give me a response within 30min for my use case.

Switching to #Mistral 3B Q4_K_S now. *fingers crossed*

Beim Chatbot F13 des DAP https://bw.schule ist jetzt die Verarbeitung personenbezogener Daten (ausgeschlossen Daten nach Art.9 DSGVO) möglich.
Als LLM wird neben OpenAi GPT-OSS 120B das schweizer Modell Swiss AI Apertus mit 70 Milliarden Parametern angeboten.

#fedilz #kunstlicheinteligenz #llm #openai #apertus #f13

You can’t have LLM as a judge, the humans need to stay in the loop, and we need to use RAG tools with legal databases to help make swifter and fairer decisions #Apertus #CHopen

You can’t have LLM as a judge, the humans need to stay in the loop, and we need to use RAG tools with legal databases to help make swifter and fairer decisions #Apertus #CHopen

You can’t have LLM as a judge, the humans need to stay in the loop, and we need to use RAG tools with legal databases to help make swifter and fairer decisions #Apertus #CHopen

You can’t have LLM as a judge, the humans need to stay in the loop, and we need to use RAG tools with legal databases to help make swifter and fairer decisions #Apertus #CHopen

You can’t have LLM as a judge, the humans need to stay in the loop, and we need to use RAG tools with legal databases to help make swifter and fairer decisions #Apertus #CHopen

The #ChatTF pilot points the way to using the data of courts and legal tool users to improve the foundation models like #Apertus being piloted in the Courts - in this case, an anonymization test #CHopen

The #ChatTF pilot points the way to using the data of courts and legal tool users to improve the foundation models like #Apertus being piloted in the Courts - in this case, an anonymization test #CHopen

The #ChatTF pilot points the way to using the data of courts and legal tool users to improve the foundation models like #Apertus being piloted in the Courts - in this case, an anonymization test #CHopen

The #ChatTF pilot points the way to using the data of courts and legal tool users to improve the foundation models like #Apertus being piloted in the Courts - in this case, an anonymization test #CHopen

The #ChatTF pilot points the way to using the data of courts and legal tool users to improve the foundation models like #Apertus being piloted in the Courts - in this case, an anonymization test #CHopen

#Apertus at the start of the Open Source AI conference in Bern organized by @CHopen #OpenSourceAI #AIconference2026 #CHopen

#Apertus at the start of the Open Source AI conference in Bern organized by @CHopen #OpenSourceAI #AIconference2026 #CHopen

#Apertus at the start of the Open Source AI conference in Bern organized by @CHopen #OpenSourceAI #AIconference2026 #CHopen

#Apertus at the start of the Open Source AI conference in Bern organized by @CHopen #OpenSourceAI #AIconference2026 #CHopen

#Apertus at the start of the Open Source AI conference in Bern organized by @CHopen #OpenSourceAI #AIconference2026 #CHopen

„Infrastructure is a necessary prerequisite [of sovereign AI], but it is the models and products that generate value.“ — Thomas Schulthess interviewed in #Netzwoche #CSCS #Apertus
https://www.netzwoche.ch/news/2026-05-06/wie-das-cscs-und-apertus-die-ki-zukunft-der-schweiz-sichern

„Infrastructure is a necessary prerequisite [of sovereign AI], but it is the models and products that generate value.“ — Thomas Schulthess interviewed in #Netzwoche #CSCS #Apertus
https://www.netzwoche.ch/news/2026-05-06/wie-das-cscs-und-apertus-die-ki-zukunft-der-schweiz-sichern

@xavierdatatech
focuses more on quality is a good approach.
Yes, one can already see that a lot is done in Switzerland. 🦾 My knowledge is that Switzerland is also the only country where an true open source multilingual LLM originated from. #Apertus I may take a interesting development, although I have read little about it lately.

Our team found new ways to put #Apertus and other locally hosted LLMs on the desktops of less-technical users #ClimateAction

While in a session with national councillors and esteemed guests, I asked #Apertus to comment on alignment with Digital Public Goods (DPGs) from the perspective of its own Charter
https://openwebui.com/posts/apertus_as_a_dpg_discussion_db13a651

Great to see that #Apertus has been highlighted as a digitally independent alternative to ChatGPT yesterday! However, this needs a quick fact-check:

- The model was trained in a local data center at CSCS, the costs of the project are almost entirely covered by Swiss public institutions;
- People from all over the world have made contributions - open source LLM development is a global community of interest;
- There are groups in several countries using or fine-tuning Apertus to improve linguistic capabilities and local knowledge;
- While Apertus can run on Amazon servers, thanks to a third-party deployment script, it runs anywhere LLMs can run;
- Quantized versions are available to fit even relatively cheap consumer grade video cards (see my blog posts for details);
- The #PublicAI web interface and Apertus 8B demo runs on AWS, however the large model is hosted by CSCS as well;
- You do not need to use Google to authenticate to Public AI, it is just a convenient way to log into #OpenWebUI - if you want another provider, please suggest it;
- Apertus is not a chatbot on its own, it is a large language model that can be deployed as part of a system to provide chat services.

I'll send this to the maintainers as well. Did I miss anything?

https://di.day/en/digital-switch-recipes/alternativen-zu-chatgpt-co#different-ways-to-cook

Great to see that #Apertus has been highlighted as a digitally independent alternative to ChatGPT yesterday! However, this needs a quick fact-check:

- The model was trained in a local data center at CSCS, the costs of the project are almost entirely covered by Swiss public institutions;
- People from all over the world have made contributions - open source LLM development is a global community of interest;
- There are groups in several countries using or fine-tuning Apertus to improve linguistic capabilities and local knowledge;
- While Apertus can run on Amazon servers, thanks to a third-party deployment script, it runs anywhere LLMs can run;
- Quantized versions are available to fit even relatively cheap consumer grade video cards (see my blog posts for details);
- The #PublicAI web interface and Apertus 8B demo runs on AWS, however the large model is hosted by CSCS as well;
- You do not need to use Google to authenticate to Public AI, it is just a convenient way to log into #OpenWebUI - if you want another provider, please suggest it;
- Apertus is not a chatbot on its own, it is a large language model that can be deployed as part of a system to provide chat services.

I'll send this to the maintainers as well. Did I miss anything?

https://di.day/en/digital-switch-recipes/alternativen-zu-chatgpt-co#different-ways-to-cook

Great to see that #Apertus has been highlighted as a digitally independent alternative to ChatGPT yesterday! However, this needs a quick fact-check:

- The model was trained in a local data center at CSCS, the costs of the project are almost entirely covered by Swiss public institutions;
- People from all over the world have made contributions - open source LLM development is a global community of interest;
- There are groups in several countries using or fine-tuning Apertus to improve linguistic capabilities and local knowledge;
- While Apertus can run on Amazon servers, thanks to a third-party deployment script, it runs anywhere LLMs can run;
- Quantized versions are available to fit even relatively cheap consumer grade video cards (see my blog posts for details);
- The #PublicAI web interface and Apertus 8B demo runs on AWS, however the large model is hosted by CSCS as well;
- You do not need to use Google to authenticate to Public AI, it is just a convenient way to log into #OpenWebUI - if you want another provider, please suggest it;
- Apertus is not a chatbot on its own, it is a large language model that can be deployed as part of a system to provide chat services.

I'll send this to the maintainers as well. Did I miss anything?

https://di.day/en/digital-switch-recipes/alternativen-zu-chatgpt-co#different-ways-to-cook

Great to see that #Apertus has been highlighted as a digitally independent alternative to ChatGPT yesterday! However, this needs a quick fact-check:

- The model was trained in a local data center at CSCS, the costs of the project are almost entirely covered by Swiss public institutions;
- People from all over the world have made contributions - open source LLM development is a global community of interest;
- There are groups in several countries using or fine-tuning Apertus to improve linguistic capabilities and local knowledge;
- While Apertus can run on Amazon servers, thanks to a third-party deployment script, it runs anywhere LLMs can run;
- Quantized versions are available to fit even relatively cheap consumer grade video cards (see my blog posts for details);
- The #PublicAI web interface and Apertus 8B demo runs on AWS, however the large model is hosted by CSCS as well;
- You do not need to use Google to authenticate to Public AI, it is just a convenient way to log into #OpenWebUI - if you want another provider, please suggest it;
- Apertus is not a chatbot on its own, it is a large language model that can be deployed as part of a system to provide chat services.

I'll send this to the maintainers as well. Did I miss anything?

https://di.day/en/digital-switch-recipes/alternativen-zu-chatgpt-co#different-ways-to-cook

Great to see that #Apertus has been highlighted as a digitally independent alternative to ChatGPT yesterday! However, this needs a quick fact-check:

- The model was trained in a local data center at CSCS, the costs of the project are almost entirely covered by Swiss public institutions;
- People from all over the world have made contributions - open source LLM development is a global community of interest;
- There are groups in several countries using or fine-tuning Apertus to improve linguistic capabilities and local knowledge;
- While Apertus can run on Amazon servers, thanks to a third-party deployment script, it runs anywhere LLMs can run;
- Quantized versions are available to fit even relatively cheap consumer grade video cards (see my blog posts for details);
- The #PublicAI web interface and Apertus 8B demo runs on AWS, however the large model is hosted by CSCS as well;
- You do not need to use Google to authenticate to Public AI, it is just a convenient way to log into #OpenWebUI - if you want another provider, please suggest it;
- Apertus is not a chatbot on its own, it is a large language model that can be deployed as part of a system to provide chat services.

I'll send this to the maintainers as well. Did I miss anything?

https://di.day/en/digital-switch-recipes/alternativen-zu-chatgpt-co#different-ways-to-cook

Just added the 100th paper to the Zotero library tracking publications and citations of interest to #Apertus https://www.zotero.org/groups/6385576/apertus

Featherless.ai now has #Apertus available - let me know if you're giving the service a go

Explore a demo of agentic development with the #PublicAI API shared in Vasco’s #Apertus #OpenHands #OpenEduDay repo: https://tinyurl.com/apertus-tutor

Explore a demo of agentic development with the #PublicAI API shared in Vasco’s #Apertus #OpenHands #OpenEduDay repo: https://tinyurl.com/apertus-tutor

Explore a demo of agentic development with the #PublicAI API shared in Vasco’s #Apertus #OpenHands #OpenEduDay repo: https://tinyurl.com/apertus-tutor

Explore a demo of agentic development with the #PublicAI API shared in Vasco’s #Apertus #OpenHands #OpenEduDay repo: https://tinyurl.com/apertus-tutor

Explore a demo of agentic development with the #PublicAI API shared in Vasco’s #Apertus #OpenHands #OpenEduDay repo: https://tinyurl.com/apertus-tutor

"You have to spend some energy and effort to see the beauty of math." -- Maryam Mirzakhani (1977-2017) #EPFL #Apertus https://en.wikipedia.org/wiki/Maryam_Mirzakhani

Some quick and sloppy poetry in Swiss-German - same prompt with #Apertus 70B on the left, 8B (feat. SEA-LION) on the right. On popular request 🙂 #Bärndüütsch

Here is an #Apertus summary of the section „Datensouveränität als moderner Heimatschutz“:

> Data sovereignty starts with artificial intelligence. It would be hubris for the Swiss Broadcasting Corporation (SRG) to attempt to develop its own Swiss Large Language Model to compete with giants like Open AI or Google. However, the SRG can provide the crucial fuel for an independent Swiss AI.

> When we use global AI models today, we get responses based on Californian values, Anglo-Saxon logic, and often a distorted view of our small-state nature. The SRG holds the unique, multilingual data treasure of this country, large enough to train national AI projects—such as the ETH's Apertus model—to understand our nuances, our four national languages, and above all, our dialects.

> Public service in 2026 means processing Switzerland's digital raw material so that our cultural identity doesn’t get washed away in the age of algorithms. The SRG must become the guardian of data sovereignty to prevent Switzerland from becoming a digital colony in the digital sphere.

In a recent NZZ article, Barnaby Skinner argues that the SRG (national broadcaster, target of a recent cost-cutting referendum) needs be a public utility, re- focus on producing high quality, open content: including for use in models like #Apertus - along with other recommendations for a democratic Swiss media landscape https://www.nzz.ch/meinung/die-architektur-der-ahnungslosigkeit-warum-der-service-public-ein-neues-betriebssystem-braucht-ld.1930447

@BlumeEvolution bin zwar lediglich Softwareentwickler, aber ich denke hier braucht es keine LLM sondern entsprechende Infrastruktur und Tools (FlowSentry, fail2ban, ...).

mastodon.social hat > 250k aktive Profile, wäre interessant ob hier bereits entsprechende Infrastruktur im Einsatz war.

Aber bzgl. Open Source LLMs: das Problem ist dass es bis auf #apertus meines Wissens keine einzige LLM gibt die vollständig und end-to-end open source ist.

(1/)

was curious about the cost of training the Swiss LLM Apertus, found an estimate in here (only an estimate? wouldn't you want to measure this as it's a thing people care about)

https://arxiv.org/pdf/2509.14233

- 7×10^24 flops (yottaflops!)
- 90 days using 4096 GPUs
- 5 GWh
- for the actual training they claim carbon neutrality as they used hydropower

#apertus

Lizzy 7B is a new UK-specific large language model launched today that performs well compared to EuroLLM 9B and Apertus 8B on key metrics. Deployable via Hugging Face or enterprise support, it offers full control, security, and data privacy for UK enterprises and public sector.

https://flower.ai/models/lizzy/

Summarized by #Apertus
#UKAI #SovereignAI #Lizzy7B #UKLLM #AIforEnterprise #ComplianceReady

Lizzy 7B is a new UK-specific large language model launched today that performs well compared to EuroLLM 9B and Apertus 8B on key metrics. Deployable via Hugging Face or enterprise support, it offers full control, security, and data privacy for UK enterprises and public sector.

https://flower.ai/models/lizzy/

Summarized by #Apertus
#UKAI #SovereignAI #Lizzy7B #UKLLM #AIforEnterprise #ComplianceReady