Summary
Google Cloud Billing claims I generated 16,600 images, but Google Cloud Monitoring shows only 200 GenerateContent requests. My application database records 9 images.
Environment
- Service: Vertex AI (aiplatform.googleapis.com)
- Models: Gemini 3.1 Flash Image, Gemini 3.0 Pro Image
- Billing Period: March 2026
Evidence
1. Google Cloud Monitoring API
Query: serviceruntime.googleapis.com/api/request_count for aiplatform.googleapis.com
| Method | Count |
|---|---|
| StreamGenerateContent | 222 |
| GenerateContent | 200 |
| Other | 50 |
| Total | 472 |
2. Google Cloud Billing
| SKU | Usage | Cost |
|---|---|---|
| Gemini 3.1 Flash Image Output (7D83-2D44-B50E) | 12,600 count | $23.68 |
| Gemini 3.0 Pro Image Output (47A8-A5A1-B26C) | 4,000 count | $15.03 |
Google Support confirmed “count” = number of images.
3. My Application Database
| Date | Images Generated |
|---|---|
| 2026-03-11 | 4 |
| 2026-03-14 | 5 |
| Total | 9 |
Discrepancy Summary
| Source | Count |
|---|---|
| My Database | 9 |
| Google Monitoring | 200 |
| Google Billing | 16,600 |
Note: My database may not capture all attempts (e.g., failed requests during testing). However, even using Google’s own Monitoring data (200 GenerateContent requests), it’s impossible to reach 16,600 images.
Billing Math Doesn’t Add Up
If “count” = images:
- 12,600 images × $0.045 (512px, lowest price) = $567
- Actual charge: $23.68
- 24x discrepancy
If “count” = tokens:
- 12,600 tokens × $60/1M = $0.76
- Actual charge: $23.68
- 31x discrepancy
No interpretation of “count” produces the actual charge.
Questions
- Why does Monitoring show 200 requests but Billing claims 16,600 images?
- How is the “count” unit calculated for image generation SKUs?
Related Issues
- Bug: Gemini 2.5 Flash Lite candidatesTokenCount underreports by ~8.5x (Topic 124448)
- Incorrect Image Token Calculation Results in Overcharging (Topic 99312)