Reducing image detail can increase the total cost of an AI request. In OpenRouter's visual-question benchmark, OpenAI's GPT-5.5 scored 65.2% with low image detail versus 79.0% with automatic detail, while average cost rose from 4.5 to 5.1 cents per question. The low-detail run produced more reasoning tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition →, outweighing its input savings.
The test used the MMMU-Pro Vision question set and a single evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → pass. It shows why an image pipeline's cost must include the model's output as well as its input: a cheaper image representation need not produce a cheaper completed answer.
Terms in this piece · Glossary
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
