Looks like a pretty significant decrease in cost compared to previous top OpenAI models. It looks like they matched Gemini 2.5 Pro’s price per input and output exactly, where O3 used to be 4x the cost vs Gemini.

I wonder if they’ve actually brought their computation costs down, or if they’re just leaning on investor capital to stay competitive price wise.

Edit: looks like they had a messed up bar graph on this slide of the presentation. Surprising to see a mistake like this slip through on such a large budget release.