Why each model has a different credit cost
It's not arbitrary. Every cost number on /models comes from a small formula tied to the actual vendor price.
Open the Models page and you’ll see numbers like “1 credit”, “2 credits”, “5 credits + 1/sec after 5s”. Those aren’t marketing tiers — each is the cost formula the backend uses to charge your account.
The shape is small: base credits, plus optional per_unit rates that multiply against parameters you choose. Flux Schnell is base 1 because a text-to-image render is fast and cheap on the vendor side. Flux Kontext Pro is base 2 because the underlying inference is roughly twice as expensive per call. WAN 2.2 Image-to-Video is base 5 plus 1 credit per extra second past the first five, because the vendor bills per-second for video too.
We seed those numbers from each vendor’s base price using a fixed credits-per-dollar conversion. The same JSONB column the Studio reads to display the cost is the column the charge path reads at submit time, so “what the page shows” and “what you pay” can’t drift.