Comment by qeternity

Sure, that's a different issue. If you prompt in a way to invoke chain of thought (e.g. what humans would do internally before answering) all of the models I just tested got it right.