Yes, but bigger models are still more capable. Models shrinking (iso-performance) just means that people will train and use more capable models with a longer context.