Operators should evaluate a model change by pricing the current and proposed configurations with the rate that applies to each configuration’s subscription, region, deployment type, and billing agreement. If provisioned throughput is involved, they should also verify throughput per PTU for the selected model and the mix of input and output tokens in a given minute. A defensible comparison keeps these rate and workload assumptions visible and consistent on both sides.
Start with the applicable rate
The cited cost guidance says to use rates tied to the relevant subscription, region, deployment type, and billing agreement. A model name alone therefore does not provide enough information to select a reliable comparison rate.
Operators should record the following for both the current and proposed configurations:
| Rate-setting field | What to verify |
|---|---|
| Subscription | The subscription associated with each configuration |
| Region | The region in which each deployment will run |
| Deployment type | The deployment method being priced |
| Billing agreement | The agreement governing the applicable rate |
If any field changes, the applicable rate should be checked again rather than carried over from the other configuration. When a field cannot be verified, the estimate should remain unresolved instead of relying on a generic rate.
Test provisioned throughput against the workload
The cited sizing guidance states that throughput per PTU depends on the model and the mix of input and output tokens in a given minute. PTU count should therefore not be treated as a workload-neutral measure that can be compared without those conditions.
For each provisioned-throughput option, operators should document:
- The selected model.
- The expected input and output token mix for a representative minute.
- The applicable throughput per PTU for that model and token mix.
- The workload and time basis used for the current and proposed estimates.
A comparison based only on the same PTU count can obscure meaningful differences between models or token mixes. The throughput value used in the estimate should remain connected to the workload it is intended to represent.
Combine the rate and throughput checks
Once the inputs have been verified, the current and proposed configurations should appear in the same comparison rather than in separate pricing and capacity exercises. Each side should show:
- Its rate-setting fields.
- The applicable rate used.
- The workload and comparison period.
- The model and input/output token mix when provisioned throughput applies.
- The throughput-per-PTU value used.
- The resulting cost estimate and the difference between the two estimates.
Keeping the workload and comparison period consistent makes the cost impact interpretable. Changing the workload in one scenario but not the other would make the resulting difference difficult to attribute to the model change.
What operators must still confirm
The cited guidance provides selection rules, not an exact fee or a fixed throughput-per-PTU value. Operators must still obtain the rate and model-specific throughput data that apply to their own configurations before stating a monetary increase or decrease.
They should also confirm that the planned subscription, region, deployment type, billing agreement, and representative token mix match the intended workload. Any service-level, contractual, or other approval requirements are not established by these two sources and require separate evidence.
Until those checks are complete, the result should be described as a cost-comparison framework rather than a final savings or overrun estimate.