The closed models aren't giving you the real thinking traces, though.

Those traces are just summaries of the reasoning.