Compare four VL models on the same image and prompt.
Compare three VL models on the same image and prompt.