I think at the moment it is not possible to combine gemini as vision model with claude as text model. Some code cleaning will be necessary to make this work.