Bedrock Provisioned Throughput with fewer than 1,000 calls a day
What does ZopNight detect here?
Amazon Bedrock bills Provisioned Throughput by the hour for as long as it exists, whether or not it gets traffic. ZopNight looks for provisioned models that get some calls but average fewer than 1,000 `Invocations` a day over 30 days, where on-demand pricing may cost less. The rule stays silent until on-demand token rates are available to price that comparison.
Signal and threshold
| Field | Value |
|---|---|
| Rule IDs | RC-1603 |
| Category | discount |
| Severity | medium |
| Metric | Invocations |
| Threshold | under 1,000 invocations per day |
| Evaluation window | 30d |
| Source | ZopNight |
| Permissions used | bedrock:ListProvisionedModelThroughputs · bedrock:GetProvisionedModelThroughput · cloudwatch:GetMetricStatistics |
Paying by the hour for capacity a few calls use
Provisioned Throughput reserves model capacity measured in model units (MUs), and you are billed hourly for it. The price depends on the model, the number of MUs and the commitment: no commitment, which you can delete at any time, or 1 or 6 months, which you cannot delete until the term ends. Billing continues until the Provisioned Throughput is deleted.
On-demand invocation charges per input and output token instead, as listed on Amazon Bedrock pricing. A reserved block that serves a trickle of requests can cost far more than the same tokens would on demand.
Measuring provisioned traffic yourself
List what is provisioned, then read its daily invocation totals from CloudWatch. Bedrock publishes
runtime metrics in the AWS/Bedrock namespace with a ModelId dimension:
aws bedrock list-provisioned-model-throughputs \ --query 'provisionedModelSummaries[].[provisionedModelName,provisionedModelArn,modelUnits,commitmentDuration]' \ --output table
aws cloudwatch get-metric-statistics --namespace AWS/Bedrock \ --metric-name Invocations --dimensions Name=ModelId,Value=PROVISIONED_MODEL_ARN \ --statistics Sum --period 86400 \ --start-time 2026-08-26T00:00:00Z --end-time 2026-09-25T00:00:00ZAdd InputTokenCount and OutputTokenCount the same way to estimate what that volume would cost at
on-demand token prices.
Signals the rule looks for
- ZopNight holds a 30-day
Invocationsseries for the provisioned model. - The series shows real activity. A provisioned model with no calls at all is handled by Bedrock Provisioned Throughput Idle.
- Average daily invocations fall below 1,000.
- A monthly cost is known for the Provisioned Throughput.
Why no finding appears today
Even when all four signals line up, the rule currently produces nothing. A saving here is the Provisioned Throughput’s monthly cost minus the projected on-demand cost of the same input and output tokens, and ZopNight does not yet have the on-demand per-token rates needed for the second half. It will not show a guessed figure or a $0 placeholder instead, so the page describes the check as it will run once those rates are available.
How the saving will be calculated
saving = Provisioned Throughput monthly cost - (input tokens x on-demand input rate + output tokens x on-demand output rate)A finding will appear only when that difference is positive.
Moving light traffic to on-demand
- Confirm the traffic is light all month, not just between launches.
- Estimate the on-demand bill from the token counts above.
- Check the model can be called on demand in your Region; some custom models are served only through Provisioned Throughput.
- Point callers at the on-demand model ID, then delete the Provisioned Throughput after any
commitment term ends:
aws bedrock delete-provisioned-model-throughput --provisioned-model-id PROVISIONED_MODEL_ARN.