Comment by GoatInGrey

Sure, but it's still useful insight to see how it performs over time. Of course, cynically, Anthropic could game the benchmark by routing this benchmark's specific prompts to an unadulterated instance of the model.