🏅 Top traders

This still seems high to me, given how long the gap was between GPT-4 and Claude 3

What am I missing?

. I tend to put this at 80% just because of the ruthless competitive dynamics at the foundation layer.

@DiegoAkel Presumably you mean other than the claude 3 family? and that a modification of that family of models would not count as a new model, but that you instead require a new architecture / size / etc.? Does it need to be an LLM, or would a text to video model count?

@Bayesian Fair enough! Will add more details to the question