Switzerland has just entered the AI race with a bold move. On Monday, the country released Apertus, an open-source large language model trained only on publicly available data.
Unlike most big tech models, Apertus comes with its code, weights, and data sources fully transparent. Developers say it’s designed to prove trustworthy AI can be built without breaking copyright rules — a sharp contrast to US rivals.
Key Takeaways
- Switzerland unveils Apertus, a fully open-source AI model.
- Trained only on public websites that allowed scraping.
- Two versions: 8B and 70B parameters, rivaling Meta’s Llama 3.
- Designed to meet strict EU copyright and AI code rules.
- Aims to set new standards for transparency in AI.
Apertus is Switzerland’s new open-source AI model, released in September 2025. Trained only on public data sources, it comes in 8B and 70B parameter sizes. Unlike proprietary systems, Apertus offers full transparency with its code, weights, and training details on HuggingFace, aiming to align with EU copyright and ethical AI standards.
Apertus: Switzerland’s Big Bet on Open AI
Switzerland has officially joined the high-stakes AI race. On Monday, developers backed by Swiss research institutions released Apertus, a fully open-source language model that directly challenges the dominance of OpenAI’s ChatGPT, Anthropic’s Claude, and Meta’s Llama.
The model’s name, Latin for “open,” reflects its philosophy: everything is transparent, from source code to training data. The developers uploaded it to HuggingFace, making Apertus one of the most openly documented large models to date.
Built on Public Data, Not Secrets
What sets Apertus apart is its training approach. According to the team, it only used publicly accessible web data and respected website opt-outs — meaning no “stealth crawling.” That design choice not only reduces copyright risks but also aligns with the European Union’s copyright directives and voluntary AI code of practice.
This directly contrasts with many US-based AI companies, which have faced lawsuits and criticism for scraping copyrighted works without consent.
Inside the Model: Two Sizes, 1,800 Languages
Apertus comes in two configurations: an 8-billion parameter version for lighter applications and a 70-billion parameter version aimed at enterprise and research use. Its multilingual reach is striking — trained on more than 1,800 languages, it seeks to serve as a globally inclusive alternative to English-heavy AI models.
For comparison, Meta’s Llama 3, released in 2024, is currently one of the most widely used open models in research and startups. Swiss developers say Apertus offers comparable performance, with the added advantage of its transparent training pipeline.
Global Implications and Industry Response
By making everything public, Switzerland is positioning Apertus as a test case for “trustworthy AI” in a regulatory-heavy environment. For European policymakers, it could become proof that innovation and compliance can co-exist.
Analysts note that Apertus may appeal to governments, research labs, and companies that need models but worry about licensing restrictions or opaque data origins.
Still, skeptics caution that transparency doesn’t guarantee adoption. Proprietary players like OpenAI and Anthropic still dominate because of their superior performance, integrations, and funding. Apertus will have to prove it can scale in real-world use cases.
Why It Matters
The launch of Apertus comes at a time when trust in AI is fragile. Lawsuits, copyright battles, and misinformation scandals have made both regulators and the public wary of “black box” models. Switzerland’s model attempts to flip the narrative by offering openness as a feature, not a liability.
If successful, it could set a new baseline for how open-source models are built — not just in Europe, but worldwide.
Conclusion
Switzerland’s Apertus is more than just another AI model. It’s a statement: that transparency, copyright compliance, and global inclusivity can define the next chapter of artificial intelligence. Whether the world adopts it at scale remains to be seen, but it marks a turning point in the AI race.