Agent run costs
Cost per agent run in the period: the estimated cost of the run’s sandboxes plus the cost of its model calls, highest total first. A sandbox belongs to a run through the run key in its metadata (run, or the CostGraph run key for sandboxes a workflow or CI job started); model calls belong to it through an AI gateway dimension named run, which pushed usage fills from the x_Run field. Sandboxes count by the day they started and model calls by their usage day. sandbox_cost_usd is estimated as in the summary, and sandboxes with no list price count in unpriced_sandboxes and add no cost. A run with only sandboxes or only model calls is listed with the other side at zero. cost_usd is sandbox_cost_usd plus token_cost_usd. Defaults to the last 30 days; a period over 400 days is rejected.