[killcoder]

> 30% more token-efficient at the same reasoning level across many tasks

But they're claiming it's more token efficient, so me switching my usage to the new model should _free up_ capacity.