Skip to Content
ExplanationCost estimation

Cost estimation

This page explains how PlainField estimates the cost of each request and why the number is an estimate, not a bill.

How it works

Every provider returns token usage in its response (prompt tokens and completion tokens). After proxying, PlainField reads those counts and multiplies them by a built-in per-model pricing table:

estimated_cost_usd =
    input_tokens  × price_per_input_token(model)
  + output_tokens × price_per_output_token(model)

There is no extra call to the provider to compute this — PlainField already has the token counts from the response body, and the prices are known constants.

The result is stored with the log entry and aggregated by /api/costs and the dashboard’s Analytics view.

Why it is an estimate

  • Pricing drift — providers change prices; the table reflects the values at the time it was last updated.
  • Unlisted models — a model missing from the pricing table contributes 0 to the cost. If total_cost_usd is unexpectedly 0, the model is probably not in the table. See Providers and models.
  • Provider-side discounts — volume discounts, committed-use rates, and promotional credits are not modelled.
  • Non-token charges — features billed outside token usage are not included.

Treat the estimate as a fast, directional signal for comparing prompts, models, and users — not as an invoice.

Keeping it accurate

Cross-check the estimate against your provider’s official billing for the periods that matter, especially if you have negotiated pricing or use models that are not in the pricing table.

Last updated on
Architecture