Price per M/tokens is also a fuzzy metric when newer models reason longer, and then burn more tokens while doing so.

Isn't 5.5 a router, though? As in, some prompts get automatically sent to a cheaper model?