In practice, models that do this won't be prioritized as much, because the economics of thinking tokens that stop by default at, say, one option plus a bit more planning (short of full alternatives) would be superior as long as billing is per-user instead of per-token. So we'll still need to play games with prompting!