I expected the cheaper model to be cheaper. It cost 8.6 more.

I'd routed the same one-word prompt to Claude Haiku and to Gemini 2.5 Flash. Flash has the lower per-token price, so this should have been an easy win. It wasn't. Flash is a thinking model: before it answered "Paris," it spent a few dozen tokens reasoning, and reasoning is billed as output. Haiku an