The prompt defaults to the free GPT LLM, but the main advantage of logging in is to leverage the most powerful models. Trying to change it inside the miniapp Browser UI just opens a new query.
Setting a model preference in settings would help streamline this.