It's beyond terrible. Like they're routing to gpt4o mini with low effort behind the scenes. Just let us pick the model and the effort.