GPT-5.2 - More vision token billing issues.
- detail:low not working on any endpoint
- different billing between Chat Completions and Responses
| model (low) | chat completions | responses |
|---|---|---|
| gpt-5-2025-08-07 | 70 | 353 |
| gpt-5.1-2025-11-13 | 70 | 353 |
| gpt-5.2-2025-12-11 | 273 | 327 |
(prior model overbilling on Responses still happening above)
| model (high) | chat completions | responses |
|---|---|---|
| gpt-5-2025-08-07 | 350 | 350 |
| gpt-5.1-2025-11-13 | 350 | 350 |
| gpt-5.2-2025-12-11 | 273 | 327 |
Likely cause
- Using “patches” algorithm; not “tiles” of any non-mini model before
- Using a 1.2x cost multiplier on Responses
Demonstration of “patches” symptom
gpt-5.2cannot see a patches-aligned checkerboard image
gpt-5.1
It’s a simple black-and-white checkerboard pattern: a grid of equally sized squares alternating between black and white in both rows and columns.
gpt-5.2
The image appears to be completely black—no visible objects, text, or details.
Documentation
- There is no update to the vision guide for pricing with algorithm: https://platform.openai.com/docs/guides/images-vision?api-mode=chat#calculating-costs
- There is no update to the pricing page with a vision calculator: https://openai.com/api/pricing/