I’m testing image questions through the OpenAI API and can’t tell how much control I have over the image the model actually sees. If I send a large screenshot or a detailed diagram, is it analyzed at its original resolution, or does the API resize or split it first?

I’m especially curious about the difference between the default setting and high-detail image input. Has anyone compared the results on small text, dense charts, or UI screenshots? I’d like to know whether choosing a detail level changes what the model can reliably read, or mainly affects token use.