DeepSeek v4 Pro is too big for such a "midrange" performance, or am I missing something?

Hi. DeepSeek v4 Pro has 1.6T parameters, probably the largest in open models, or at least one of the largest. Yet it's not the best/most performance open model, considering a wide variety of definitions of "best". Indeed, in most cases, it is not the second best, third best, or fourth best either. G