Just on Claude Code, I didn't notice any performance difference from Sonnet 4.5 but if it's cheaper then that's pretty big! And it kinda confuses the original idea that Sonnet is the well rounded middle option and Opus is the sophisticated high end option.
It does, but it also maps to the human world: Tokens/Time cost money. If either is well spent, then you save money. Thus, paying an expert ends up costing less than hiring a novice, who might cost less per hour, but takes more hours to complete the task, if they can do it at all.
It's both kinda neat and irritating, how many parallels there are between this AI paradigm and what we do.
