Usage and model prices

Settings → Usage answers "where is the money going" without exporting anything.

What it shows

• Totals for the last 7, 30 and 90 days, plus a projection for the end of the month.
• A per-day cost chart — the fastest way to spot the day something ran away.
• By model, filterable, so an expensive model in a routine flow stands out.
• By person, so you know whose work is driving spend. Runs that predate this attribution appear as Unattributed.
• Per-workflow cost, success rate and average latency per model — the operational view of whether a flow is worth what it costs.

Every figure is the same estimate shown on runs, booked per item as work completes.

Model prices

Below the usage panels, Model prices lists what each model costs in that display.

• Built-in and auto-synced prices maintain themselves. OpenRouter's public price feed is re-read every few hours, and the built-in catalogue is cross-checked against it so a price change is noticed rather than silently mispriced.
• An admin can add a manual override when a real bill disagrees with the list. Overrides win, and take effect within a few minutes across the app and the workers — no redeploy.
• Auto-synced and built-in rows are read-only; you override them rather than editing them.

Reading it honestly

• These are estimates of what your providers will bill you. Your invoice is the source of truth. OpenRouter-routed runs are the exception: those book the exact amount charged.
• $0.00 on a real generation means the model is missing a price, not that the call was free. Add an override.
• Some per-call figures are estimates rather than metered (fal.ai background removal, for one) — good enough to budget against, not to reconcile with.

Worth doing once

Compare a week of FlowArray's numbers against your actual provider dashboards. If they agree, you can trust the projection and set a daily spend limit with confidence. If they do not, the override field is right there.