Decision cost
What would the decisions cost?
The argument for a cheap judgment layer is an arithmetic one: at $0.042 per million input tokens, judging every item in a stream stops being a budget question. This works that arithmetic out with your numbers, so you can see the size of it rather than take it on faith.
Four inputs, all editable, all multiplied in the browser. Nothing is sent anywhere and nothing is stored.
How many times a day this judgment gets made — every item in the stream, not just the ones a person sees today.
The state you send with the question: the text, the option list, and any evidence fields. Output is not metered, so only the input side is counted here.
$0.042 per million input tokens is a price TypeSafe publishes, not a bill anyone here has paid. Edit it to whatever you are actually quoted. See the claim · listing
Whatever one decision costs you today — a generative model call, a rules engine, or a person’s minute priced as money. Leave it at zero to skip the comparison.
One decision at today’s route pays for 36× decisions at the price entered — $58.32 a day of difference at 50,000 decisions.
Track cost per completed task, not cost per call. A cheap wrong decision can cost far more than the call that made it — in retries, in the work downstream of it, and in whatever it took to notice.
Arithmetic only. Nothing is sent anywhere, and nothing here measures how many tokens your questions would actually need, or how often the answers would be right.
How to keep the token count downHow to read the number
The price is a vendor claim
TypeSafe publishes a low input-token price and no metered output charge. Nobody here has verified an invoice against it, and a published list price is not the price you are quoted. Edit the field.
Token counts are yours to measure
The default of 800 input tokens is a placeholder. Your real number depends on how much state you send, how long the option list is, and how much evidence rides along with the question. Measure it on your own traffic.
Volume is the variable that moves
Cheap calls change what is worth judging, not just what an existing judgment costs. The interesting question is usually how many things you would start deciding, not how much the current ones get cheaper.
Latency and accuracy are not priced here
This page multiplies four numbers. It says nothing about how fast an answer comes back, how often it is right, or what a wrong one costs downstream.