The race for scale has ended, the race for trust has begun

Next month, a Swiss LLM trained at the Swiss National Supercomputing Centre (CSCS) arrives with a capability NO other model has.

It might reduce the language barriers to global commerce.

This model, a joint effort by ETH Zurich and EPFL, was trained with over 16 trillion tokens across more than 1,500 languages. It's engineered for a level of translation accuracy that current systems cannot reliably offer.

Next to the multilingual capacity there is another big advantage: its transparency.

The "open-source" models from Meta, DeepSeek, and OpenAI are not built on fully open data and they don't allow you to look at the data they are trained with. This Swiss model is the first major LLM trained exclusively on publicly available sources, and its entire dataset will be disclosed.

This means biases can be identified and mitigated for each specific use case, a foundational step for building transparent, fair, enterprise-grade AI.

European companies, the German Mittelstand and regulated professions no longer have to trust a handful of billionaires across the Atlantic and can build their own AI systems that they and their clients actually can trust.

What might become possible with this sovereign model for the first time?

-Sovereign Global Teams. An engineering team here in Munich can communicate with a supplier in Vietnam without language friction. Project updates and technical queries can be exchanged in each team's native language, processed securely within their own infrastructure.

-Compliant Cross-Border Intelligence. A German law firm can analyze a Spanish-language contract, or a French doctor can review a Polish medical history, without ever challenging client confidentiality and data residency.

-True Market Understanding. You can analyze customer feedback from dozens of countries in its original language, gaining nuance that is lost in generic translation, all while the raw data stays under your control.

This model is a leading indicator of where the serious enterprise market is heading.

The first era of generative AI was a race for scale, measured in parameter counts and dominated by a few massive, generalist models. GPT-5 showed that this era is over.

The next era is a race for trust, measured in verifiability, transparency, and jurisdictional security. We are witnessing the emergence of the "workhorse" AI category: specialized models that perform high-value work with maximum efficiency inside sovereign systems.

The race for scale has ended. The race for trust has begun. That is the new benchmark.