I'd assume the closed weight models are all working on shrinking their parameter counts anyways. They too benefit from smaller models. It'd be foolish for these SOTA labs to not be working at reducing parameter counts.