Turn

One inference call, the unit `fill.maxTurns` counts.

The budget counts calls rather than money, because every model makes calls and each one can be counted. One page is one turn, and a cached page spends none. Converting turns to a cost is your provider’s rate card.