Pricing and exact billing
Usage meters, caching, markup, postpaid charges and reconciliation.
AI has no free generation allowance. No funds are reserved before provider submission. Verified actual usage is charged afterward, and the balance may go negative. Customer monthly caps stop new work when settled usage reaches the cap; in-flight work may finish above it.
Markup and margin
The default 10% markup charges $1.10 for a $1.00 reviewed billing basis. Most routes use verified provider cost as that basis. This differs from 10% gross margin, which divides the billing basis by 0.90. The operator chooses the mode; the immutable price snapshot records it.
Provider cost uses exact decimal arithmetic at 18 decimal places. Markup is applied to the exact customer billing basis, then rounded up once to the ledger's micro-USD precision. Do not independently round each token category, use a binary floating-point subtotal, or calculate invoices in the browser.
A supplier contract can differ from the customer tariff. Coconut GPT Image routes use an operator-confirmed flat supplier price per returned image and OpenAI token meters for customer billing. Cloud applies markup to the token subtotal and reports both amounts separately. This explicit policy has no hidden minimum, even when the customer subtotal is below supplier cost. The supplier tariff, customer meters and margin are frozen on each request.
Meter coverage
Price cards must cover uncached input, cached reads, cache writes, output, reasoning, image/audio tokens, seconds and other usage reported by the route. Cached tokens are not automatically free. Overlapping usage totals must be interpreted by the reviewed provider adapter so they are not charged twice.
OpenRouter routes use its reported usage.cost, representing the amount charged to Cloud's upstream account. Other routes use reviewed meters. No missing price or usage is silently treated as zero.
Read the billing record
GET /v1/ai/tasks/{id} returns billing.provider_cost_usd, the customer retail_basis_usd, and charged_micros separately, plus the price version and breakdown. reserved_micros is "0" for new requests. Treat decimal money and integer micros as exact values. One dollar equals 1,000,000 micro-USD.
Verified completion charges actual usage. A timeout, interrupted stream, missing usage or unknown outcome keeps billing pending and may report reconciliation_required. Never create a fresh idempotency key to recover an uncertain request. See billing accounts and errors.