Have you read the chain of thought output from reasoning models? That's not what it does.