A fully open, multilingual foundation model built in Switzerland for sovereign AI.
Last verified · 2026-06-22 · by Moe Ameen
Apertus is a fully open large language model developed by the Swiss AI Initiative — a collaboration between EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS). It was released on September 2, 2025. The name is Latin for "open," which is the whole point: not just the weights but the training data, source code, training recipes, and alignment principles are published and documented so the model can be inspected and reproduced. That goes further than "open weights" releases that ship a checkpoint without the data or method behind it.
It comes in two sizes, an 8-billion-parameter model and a 70-billion-parameter model, each available as a pretrained base and an instruction-tuned variant. Apertus was trained on roughly 15 trillion tokens spanning more than 1,000 languages, with about 40% of the data outside English — including languages usually left out of large models, such as Swiss German and Romansh. It supports a long context window (65,536 tokens) and is released under the permissive Apache 2.0 license, which allows research, education, and commercial use.
The training data was assembled with EU AI Act transparency obligations and Swiss copyright and data-protection law in mind: the corpus uses only publicly available data, filters out content from sites that opted out (even retroactively), and strips personal data before training. That compliance story, plus full reproducibility, is why Apertus is framed as a building block for "sovereign AI" — AI that public institutions, governments, and companies can run and audit themselves rather than renting from a closed provider.
One thing to be clear about: Apertus is a text and reasoning model. It writes, translates, summarizes, and answers — it does not generate images, video, or audio. You can download it from Hugging Face (the swiss-ai org), run it locally through tools like LM Studio, or reach it through partners such as Swisscom and the Public AI Inference Utility. Like any LLM, it can produce inaccurate or biased text, so its output needs checking before it ships.
Apertus is strongest where most models are weakest: multilingual and sovereign text. If your audience spans several languages, you can draft a caption, hook, or script in each of them with one open model — and because the weights are downloadable under Apache 2.0, a team that needs copy drafted on its own servers (no data leaving the building) can run Apertus locally. What Apertus cannot do is turn that text into a post. It has no image, video, captioning, or publishing layer at all.
That gap is exactly what Kompozy fills. Take an Apertus-drafted concept and Kompozy generates the media the model can't: persona and avatar video with native captions, Photo Posts and face-locked Persona Photos, multi-slide Carousels and Quote Graphics rendered pixel-exact through HyperFrames, plus blogs and newsletters. Then it publishes — fanning each piece across the nine supported platforms (Instagram, Facebook, TikTok, YouTube, LinkedIn, X, Pinterest, Threads) plus email and blog, on a schedule, on autopilot, with a per-post review pipeline. So a multilingual idea drafted in Apertus becomes a localized week of finished, on-brand posts per market instead of a wall of raw text. Apertus owns the open, sovereign words; Kompozy owns the media, the brand styling, and the publish.
Apertus is a fully open, multilingual large language model built by the Swiss AI Initiative (EPFL, ETH Zurich, and CSCS) and released on September 2, 2025. It comes in 8B and 70B sizes, was trained on about 15 trillion tokens across more than 1,000 languages, and ships its weights, training data, code, and methods openly under the Apache 2.0 license.
Yes. Apertus is released under the Apache 2.0 license, which permits research, education, and commercial use. You can download the weights from Hugging Face and run the model yourself, or access it through partners like Swisscom and the Public AI Inference Utility.
No. Apertus is a text and reasoning model — it writes, translates, and summarizes but does not produce images, video, or audio. To turn its text into posts, you pair it with a generation and publishing engine like Kompozy, which renders the media and publishes across platforms.
Because its full pipeline is open and reproducible — weights, training data, code, and alignment — and it was built with EU AI Act transparency and Swiss data-protection law in mind. That lets governments, public institutions, and companies run and audit the model on their own infrastructure instead of depending on a closed provider.
Apertus writes the copy but does not publish. Bring an Apertus-drafted script or caption into Kompozy to generate video, carousels, quote cards, blogs, or newsletters in your brand voice, then schedule and publish the set across TikTok, Reels, Shorts, X, LinkedIn, and more from one queue.