Open models
This chapter takes about 20 minutes in the workshop.
What you will be able to do
- Place any model on the four-tier openness ladder.
- Say what is open and what is not in Apertus.
- Look up what a model was trained on, and what you cannot find out.
- Name the three reasons “open” matters for a library.
The workshop uses an open model on purpose. For a library, “open” matters for three reasons.
- Provenance: what did the model read, and can you check?
- Reproducibility: can you get the same answer next year, with the same model file?
- Vendor independence: if a provider changes its terms or its prices, can you switch to another one?
The reasons from earlier editions of this workshop still apply: privacy (your data stays on your machine), cost (no per-token bill), offline use, and control over which version you run.
SWIB’s own draft community AI policy (October 2026) takes the same view. It prefers locally hosted, open models, and it says a model whose training data is not available should not be called open source.
“Open” is a spectrum, not a yes or no. The first page gives you a ladder to place any model on.