June 21, 2026
Open AI or open season?
Apertus – Open Foundation Model for Sovereign AI
Europe’s big “open AI” launch is splitting the crowd between hope, side-eye, and roast mode
TLDR: Apertus unveiled an openly documented AI model meant to meet Europe’s strict rules on privacy, copyright, and transparency while supporting over 1,000 languages. Commenters instantly split into camps: some called it old, weak, or untrustworthy, while others said it’s already good enough for practical software use.
Apertus is pitching itself as the clean, transparent, rule-following AI: fully documented, reproducible, built to respect privacy choices, and designed with Europe’s new AI rules in mind. On paper, it sounds like the wholesome answer to the usual black-box chatbot chaos. It also says it can compete with leading open models at similar sizes and speak more than 1,000 languages, which is the kind of stat that screams “global ambition.” But in the comments, the real action is less polite press release and more open-season skepticism.
The sharpest reaction came from users who basically said: nice promises, but we’ve heard this before. One commenter flat-out called the earlier version “pretty bad” and even accused its copyright-friendly claims of not holding up in testing, ending with the brutal verdict: “completely useless.” Ouch. Another zoomed in on whether this is even a fresh model at all, suggesting the instruct versions look like an older Llama fine-tune and asking, in so many words, where’s the actual progress? That sparked a mini-doom spiral where one user declared their “last hope for sovereign AI” might now be Chinese open models instead.
But not everyone came with pitchforks. One practical user stepped in with a calmer take: no, it’s not ready to be your autonomous robot employee, but for search-and-answer software it’s “pretty competent” and useful today. Meanwhile, another commenter arrived with the classic comment-section power move: naming rival open models and basically turning the thread into a who’s really the most open? showdown. So yes, Apertus launched a model — but the community launched a trust issue.
Key Points
- •Apertus is presented as an open foundation model for sovereign AI.
- •The article says its training data, code, weights, methods, and alignment principles are documented and reproducible.
- •The model is described as being built to meet EU AI Act requirements.
- •The article claims Apertus respects opt-outs, removes PII, and includes measures to prevent memorization.
- •Apertus is said to be competitive with top open models at 8B and 70B parameters and trained on more than 1,000 languages.