[changoplatanero]

Both are true. One spent $400 in compute and the other one spent a lot more.

[masterjack]

Exactly. And presumably had a more sophisticated harness around the model, longer reasoning chains, best of N, self judging, etc