01 / The Model
Open-weight model
A model whose parameters are published, so you can run it on your own infrastructure. The alternative is a closed model you can only reach through its provider.
How you’ll hear it
“Can we keep everything on our own servers?”
“With an open-weight model, yes. We would run it ourselves.”
Avoid
Calling open-weight models open source. The weights are available; the training data and method usually are not.
The longer version
Open-weight models let an organisation run AI on its own hardware or chosen cloud, which helps with data control, cost at scale and customisation. The price is operational: you take on hosting, updates and security. Closed models are easier to start with and often ahead at the frontier.