Metering & billing

How requests turn into credits and how credits turn into USDC on Base.

Every billed request produces two credit rows: providerCredits (paid out) and operatorCredits (network revenue). Credits accumulate through the week; each Monday 00:00 UTC the coordinator sweeps and marks payout-ready.

Credit formula

grossCredits    = ((promptTokens × promptRate) + (completionTokens × completionRate))
                  × globalMultiplier
                  ÷ (1M × 1M)

providerCredits = grossCredits × providerShareBps ÷ 10,000
operatorCredits = grossCredits − providerCredits

Both values use banker’s rounding — ties round to the nearest even quotient, eliminating systematic bias across large volumes of small requests. Constants:

Symbol	Default	Meaning
`tokensPerMillion`	1,000,000	Rate card denominator
`globalMultiplierDenom`	1,000,000	PPM denominator
`providerShareDenom`	10,000	Basis-points denominator
`maxBillableTokens`	10,000,000	Per-field overflow guard

Default rate card:

Token type	Rate
Prompt	500,000 credits per 1M tokens
Completion	1,000,000 credits per 1M tokens

Default provider share: 90%. Default global multiplier: 1.0. The operator can override rates per model or add model-specific entries. Faulted requests (breaker_qualifying or null_usage_error) produce zero credits.

Prefix-cache billing

For sticky-routed requests where the provider reuses its warm KV-cache, usage.cached_prompt_tokens reports the count of prefix-cache tokens. These are billed at the per-model prompt_cache_hit_rate_per_mtok rate instead of the full promptRate. The OpenAI usage object is extended with one field:

{
  "prompt_tokens": 1500,
  "completion_tokens": 300,
  "total_tokens": 1800,
  "cached_prompt_tokens": 1200
}

cached_prompt_tokens is always present; 0 when there was no cache reuse.

Verified model settlement

For settlement-capable traffic (v0.3+ receipts), the coordinator receipt verdict drives the gateway’s buyer debit/refund decision on the money path.

Valid receipt → buyer debited, provider credited, settlement proceeds.
Missing / malformed / unverifiable receipt → settled as FaultBreakerQualifying with zero provider credits. Buyer is not debited.
Late receipt. For settlement-capable traffic a provider receipt must arrive before the coordinator’s pending_deadline_seconds window closes. Receipts arriving after the deadline are non-settling and non-recoverable.

Full detail: Operations → Verified model settlement.

Settlement cadence

Cadence: every 7 days at Monday 00:00 UTC.
Minimum payout threshold: 500,000 credits per provider. Below the threshold, credits roll to the next window.
Rail: USDC on Base.

StartWeeklySettlement schedules the sweep to fire at the next Monday 00:00 UTC and repeat weekly. settlement.job_enabled must be true (default) for the job to execute.

Endpoints

Endpoint	Purpose
`GET /v1/rate-card`	Live per-model rate card, priced against the OpenRouter frontier
`GET /v1/usage`	Buyer’s usage aggregated by day / model
`GET /providers/{provider_id}/earnings`	Provider balance, current window, last payout ready

Provider guide: Payments & payouts.

Litepaper

Getting started

API reference

Network & trust

Guides

Agentic tooling

CLI reference

Operations

Ledger

Metering & billing

Metering & billing

Credit formula

Prefix-cache billing

Verified model settlement

Settlement cadence

Endpoints

​Metering & billing

​Credit formula

​Prefix-cache billing

​Verified model settlement

​Settlement cadence

​Endpoints

Metering & billing

Credit formula

Prefix-cache billing

Verified model settlement

Settlement cadence

Endpoints