Does anyone have a benchmark that clearly distinguishes the larger models? I would think that the high parameter count models would have capabilities distinct f

Does anyone have a benchmark that clearly distinguishes the larger models? I would think that the high parameter count models would have capabilities distinct from the smaller ones, that would easily be read out. For example, Opus 4 has apparently memorized many books. If you ask it just right (to get around the infuriating copyright controls), it will complete a paragraph from The Wealth of Nations or Aristotle’s Nicomachean Ethics in Ancient Greek. That cannot be possible on a smaller model that needs to compress more.