Then your comparing to a level of GPT 5.6 High, what is 50% cheaper then Opus Medium for the same intelligence / score.

You see the issue, if you try to scale effort down, you also need to compare how other competing models compare.