Assess capability–brain coupling along a single model training trajectory

Determine how capability–brain coupling changes when a single omni model is followed throughout its training trajectory, without relying on the shared axis of a seven-model comparison panel.

Background

The paper reports brain–model correspondence across seven omni models and finds that inter-model differences are concentrated along a low-dimensional shared axis. The authors identify following one base model during training as a way to estimate capability–brain coupling without that panel-level confound, leaving the trajectory-level relationship unresolved.