Comment by kzrdude

I think that's called test-time scaling i.e using more tokens at infer time to squeeze out higher model performance. That's must be part of the explanation for good benchmark results.