Yes they do, if the model size / vram requirement keeps shrinking for a given performance target, like has been happening, then it gets cheaper to run X level o

Yes they do, if the model size / vram requirement keeps shrinking for a given performance target, like has been happening, then it gets cheaper to run X level of model.