You have LLMs but you also need to model actual intelligence, not its derivative. Reasoning models are not it.