Tracking Bedrock Costs with the Cost Explorer API

The RECORD_TYPE Filter

AWS Cost Explorer returns several record types by default: Usage, Tax, Credit, and Refund. Without filtering, credits inflate the apparent cost. Always filter to RECORD_TYPE = Usage for gross spend figures.

def sync_costs(start_date: str, end_date: str):
    client = boto3.client("ce", region_name="us-east-1", ...)

    response = client.get_cost_and_usage(
        TimePeriod={"Start": start_date, "End": end_date},
        Granularity="DAILY",
        Filter={
            "Dimensions": {
                "Key":    "RECORD_TYPE",
                "Values": ["Usage"],
            }
        },
        GroupBy=[{"Type": "DIMENSION", "Key": "SERVICE"}],
        Metrics=["UnblendedCost"],
    )
    for result in response["ResultsByTime"]:
        date = result["TimePeriod"]["Start"]
        for group in result["Groups"]:
            service = group["Keys"][0]
            amount  = Decimal(group["Metrics"]["UnblendedCost"]["Amount"])
            DailyCost.objects.update_or_create(
                date=date, service=service,
                defaults={"amount_usd": amount}
            )

Bedrock Token Logging

The cost tracker also maintains a local log of Bedrock invocations with estimated costs, so you can see which features are expensive without waiting for the Cost Explorer sync (which has a 24–48 hour lag).

# cost_tracker/tracker.py
PRICE_PER_1K_IN  = Decimal("0.00006")   # Nova Lite input tokens
PRICE_PER_1K_OUT = Decimal("0.00024")   # Nova Lite output tokens
PRICE_PER_IMAGE  = Decimal("0.04")      # Stable Image Core per image

def log_bedrock(model_id, operation, caller, tokens_in=0, tokens_out=0, images=0):
    cost = (
        Decimal(tokens_in)  / 1000 * PRICE_PER_1K_IN  +
        Decimal(tokens_out) / 1000 * PRICE_PER_1K_OUT +
        Decimal(images) * PRICE_PER_IMAGE
    )
    BedrockLog.objects.create(
        model_id=model_id, operation=operation, caller=caller,
        tokens_in=tokens_in, tokens_out=tokens_out, images=images,
        estimated_cost_usd=cost,
    )