Three models.
One machine.

Pick a hardware class and up to three models. The engine resolves the best quantization for each and grades the fit — read them down the columns like a spec sheet.