As far as analysis of social situations, political factions, incentives, strategic landscapes, social theory of mind, and the influence of cognitive traits and personality in decision making, I still feel that OpenAI's O3 was the best model overall. Even Claude 4.7 doesn't give me the same depth and richness, and isn't willing to dissect directly and without shame the subagentic structure of people involved in a certain decision (say, a bill or a topic discussed in a cabinet meeting). O3 felt uniquely genius at this, though perhaps the later relative recess of capabilities is intentional (or just not focused on during training).