@haier
With love, from Switzerland!๐จ๐ญ
This is the title of the announcement of Apertus a new Swiss open-source LLM and why it's exciting for me?
Think... most frontier AI models today are built in the same places by the same groups of people shaped by the same cultural assumptions.
That is a a huge limitation for diversity and inclusivity.
Apertus it's trained on more than 1,000 languages and released under an open licence. This means that it's not just another English-first model with some multilingual additions, but it's designed from the ground up to serve users and communities across linguistic boundaries.
This matters for diversity in three ways:
1๏ธโฃ Diversity of language. Native training across 1,000+ languages changes who AI can serve.
2๏ธโฃ Diversity of governance. Apache 2.0 and open datasets let anyone study, adapt and audit the model.
3๏ธโฃ Diversity of perspective. A model built in a different cultural, legal and ethical environment produces different priorities.
Apertus might not beat GPT-5 on benchmarks but it brings something the frontier-model race is missing: diversity of who builds AI and diversity of who it's built for.