Comment by animal-husband

That is what was observed - o1 family models performed the “cheat”, non-reasoning models didn’t.