cupel every step, and what it cost

Published runs

Every run here is published in full: the report, each step, the model that ran it, and what that step actually cost. The numbers are measured from the provider's own token counts, not estimated.

Nothing published yet.

Runs appear here once they have been run on real ideas and published. Publishing is the point rather than a side effect: the per-step cost is simultaneously the proof that the numbers are honest, the marketing, and the operator's own margin measurement.