All open model that have reasoning seem to be doing it in text tokens. Is there any indication that closed models are approaching this somehow fundamentally dif

All open model that have reasoning seem to be doing it in text tokens. Is there any indication that closed models are approaching this somehow fundamentally differently?