September 8, 2026
Pricing 120 models with one formula
Every provider prices differently: per image, per second, per token, per megapixel. You get one number. Here is how those two things get reconciled.
One provider charges per generated image. Another per second of video. Another per thousand tokens, and another by output resolution. You should never have to hold five pricing models in your head, so you hold one: credits.
One conversion, applied everywhere
A credit is a fixed amount of money. Every provider price is converted through that same constant, so a model's credit cost is derived and not negotiated. Nobody on our side picks a number that feels about right for a new model.
Where it goes wrong
Two traps have caught us more than once, both from models that don't price the way their neighbours do:
| Trap | What happens |
|---|---|
| Flat-priced image models | Priced per image whatever the size, so cost that doesn't scale with resolution is easy to model as if it did |
| Duration-selected video | You pick 5s or 10s, and billing that ignores the choice is wrong by exactly a factor of two |
What keeps it honest
- One source of truth. The estimate in the composer, the number the agent asks you to approve, and the charge on your result all read the same table. They can't disagree.
- Adding a model is a checklist, not a judgement call. A new model touches a fixed set of files, and missing one of them is how a model ends up free or unusable.
- Margins get audited. We re-check every model against current provider pricing periodically, because provider prices move and a stale conversion quietly turns into either a loss or a rip-off.
The visible result of all this is deeply boring, which is the point: the number you saw before you clicked is the number you were charged.