Use exact when…
- The cost per call is deterministic (a lookup, a fixed record, a canned response).
- The response size is bounded.
- You want buyers to be able to budget precisely.
exact unless you have a reason not to.
Use upto when…
- Cost depends on input (LLM tokens, transcription minutes, translation characters).
- Response size is unbounded and you charge by volume.
- You cannot predict compute cost until the job runs.
Common mistakes
- Using
uptowith a huge ceiling for a cheap deterministic call. Buyers set per-call caps; a ceiling of 0.001 call makes you unfilterable and hurts rank. - Using
exactfor LLM completions and quoting worst-case tokens. Buyers overpay every call and go elsewhere. - Forgetting to declare
pricing. Buyers cannot decide whether youruptoceiling is reasonable without knowing the unit rate.
Mixed offerings
A single endpoint can return multiplePaymentRequirements and let the client pick. Use this sparingly — two options is manageable, five is confusing.