Metering

Metering is the part of the request that turns provider output into a bill. Compose does it after execution, using the resolved catalog row and the usage or media evidence returned by the adapter. For text models, that usually means tokens. For media models, it can mean generated images, audio seconds, video seconds, or provider-specific units from the catalog pricing table.

Amounts

All settlement amounts are USDC atomic units. USDC has 6 decimals:

1 USDC = 1000000

The final amount is:

provider amount + platform fee

The API applies a 1% platform fee on top of the raw amount. Our fees are the lowest in the space, minimum 5.5x times lower than other aggregators (e.g., OpenRouter charges 5.5%).

Meter subject

Meter subjects use the resolved provider and public model ID:

{provider}:{modelId}

Examples:

azure:gpt-5.5
openai:openai/gpt-5.5
cloudflare:@cf/meta/llama-3.3-70b-instruct-fp8-fast

The public model ID stays in the subject so receipts match what the client requested.

Line items

A metered quote contains one or more line items:

{
  "subject": "azure:gpt-5.5",
  "lineItems": [
    {
      "key": "input",
      "unit": "usd_per_1m_input_tokens",
      "quantity": 1000,
      "unitPriceUsd": 1.25,
      "amountWei": "1250"
    }
  ],
  "providerAmountWei": "1250",
  "platformFeeWei": "13",
  "finalAmountWei": "1263"
}

Units are normalized before settlement. The metering code supports per-unit pricing, per-1k and per-1m units, and time-based units such as seconds.

Authorization and settlement

Compose separates authorization from settlement:

Phase	What happens
Authorization	The request reserves an exact amount or an `upto` cap.
Execution	The selected adapter runs the provider call.
Settlement	The API extracts authoritative usage/media evidence, prices it, and settles the final amount.
Abort	If execution fails before settlement, the prepared payment is released.

Audio requests can be quoted exactly up front when request-side billing metrics are known. Usage-priced text, image, and video calls usually use a budget cap and settle after output.

Quote a model manually

curl https://api.compose.market/api/payments/meter/model \
  -H "Content-Type: application/json" \
  -d '{
    "modelId": "gpt-5.5",
    "provider": "azure",
    "modality": "text",
    "usage": {
      "promptTokens": 1000,
      "completionTokens": 250,
      "totalTokens": 1250
    }
  }'

Use this endpoint when you need to inspect how a catalog row prices a known usage record.

Overview

External Use

Families

Modalities

Guides

Amounts

Meter subject

Line items

Authorization and settlement

Quote a model manually

​Amounts

​Meter subject

​Line items

​Authorization and settlement

​Quote a model manually

​Related

Amounts

Meter subject

Line items

Authorization and settlement

Quote a model manually

Related