Stop the overbilling of input images to gpt-4o-mini!

API Feedback
,
1

It’s been a year of inflated prices sending input images to gpt-4o-mini for vision, and needing to tell people not to use this model, because it actually costs more per image than the full gpt-4o model.

The price inflation was set against the double-price 2024-05-13 version of gpt-4o (to make sure nobody got cheap pictures from anywhere?) Now, inflating the costs by 33x the token input is just ridiculous, and doesn’t correspond to any good-faith transparency in computational costs.

The image compares the processing costs and token usage of three different GPT models (gpt-4.1, gpt-4o, and gpt-4o-mini) for the same image input, showing their respective prices, model versions, and input token counts. (Captioned by AI)
The image compares the processing costs and token usage of three different GPT models (gpt-4.1, gpt-4o, and gpt-4o-mini) for the same image input, showing their respective prices, model versions, and input token counts. (Captioned by AI)528×544 13.6 KB

If there’s a compute scarcity, you’d want to make a smaller AI model not be a bad idea for developers to use.

2 Likes