Hi Gemini API team,
I’m trying to understand the correct way to reconcile Gemini 3 Pro Image usage between API usageMetadata and Cloud Billing SKU reports.
We compared Cloud Billing Reports grouped by SKU against our application-side aggregation of API usageMetadata for the same usage window.
Observed behavior:
- Gemini 3.1 Flash Image matches exactly between Cloud Billing SKU usage and API usageMetadata aggregation.
- Gemini 3 Pro Image does not match:
- Image output billing SKU usage is about 3.2x higher than API usageMetadata image output token aggregation.
- Text output, including thinking tokens, is about 2.0x higher than API usageMetadata text/thinking output aggregation.
- Input SKUs are very close, so this does not look like a large number of missing requests.
We already include thinking tokens in output accounting.
What is the correct way to reconcile Gemini 3 Pro Image usage between API usageMetadata and Cloud Billing SKU reports?
If Cloud Billing applies additional metering rules that are not directly reflected in usageMetadata, which fields or formulas should we use to calculate the billable SKU usage correctly?