In this case the limitation is the compute. Very few people have the compute required for AI/LLMs locally or for free (comparable to the performance of Claude). So yes, there are plenty of Open Source models that can be used locally but you need to invest in hardware to make that happen and especially if you want the quality that is available from the commercial offerings.

Not to speak of the training of those models. It's all there to make it possible to do this locally however where's the hardware? AWS? Google? There are hidden costs of the Open Source model in this case.