Can't wait to hear how it breaks all the benchmarks but have any differences be entirely imperceivable in practice.

In my opinion most Anthropic models are the opposite, scoring well on benchmarks but not always way on top, but quietly excellent when you actually try to use them for stuff.