Models with reasoning > Instant have an inner monologue and when asked why they did something they can deduce based on that monologue. I have asked things like "what steps did you take" and "what was your reasoning" and the answer matched the thinking output of the model.