Open weights
A model you can run, inspect, host, and adapt yourself.
An open weights model publishes its learned parameters so teams can run, inspect, or fine-tune it themselves.
What it means
Weights are the learned numeric parameters inside a model. When they are open, a team can download the model and run it outside the original provider API.
Why product teams care
Open models are attractive when privacy, latency, unit economics, or customization matter more than buying every token from a frontier API.
Understudy angle
Understudy can compare an open model route against the expensive baseline on the same workload and only route traffic when the eval clears.
Open weights create deployment control, but the license, data story, and serving cost still matter.
Compare one open model and one frontier model on the same held-out examples.