Seems reasonable. You could just switch the model after the code is written, when you need to ask questions. You’ll lose your token cache, but that’s it.