Why can we say fable 5.2 is better than 5.1 if it needs double the tokens to complete the task? Somehow shouldn't it be compared at same token usage level to see which one is smarter? Why is not the case?