Switzerland's Public Institutions Built an Open LLM Without Corporate Help

Frequently asked

What does publicly funded mean for long-term access to Apertus weights?

Public institutional funding removes the commercial incentive to restrict access later. Unlike corporate open-weight releases — where license changes, API gating, or acquisition can alter terms retroactively — a model built by ETH Zurich, EPFL, and a national supercomputing centre is accountable to public mandates. The weights stay open because there is no business model that benefits from closing them.

Why does it matter that Apertus is multilingual at scale?

Most frontier open models are English-dominant, with multilingual capability added as a secondary objective. Switzerland's linguistic plurality — German, French, Italian, Romansh — made multilingual parity a design requirement, not an afterthought. That makes Apertus a more rigorous test case for whether public institutions can match corporate models on the hardest capability dimensions, not just the easiest.

Can other countries replicate the Swiss model for sovereign open AI?

The Swiss model requires national supercomputing infrastructure and coordinated commitment across multiple research universities — both of which exist in Germany, France, Japan, and the UK. The barrier is political will, not technical capability. Countries with equivalent compute infrastructure and research universities can build sovereign open models; Switzerland's release proves the institutional coordination is achievable.

More on this wire