I think this task could easily be handled by an on device quantized model, I’m sure these companies will start thinking about their costs, right? (Maybe not though)