“Recent models like Gemini 3, GPT-5, and Opus are significantly better at expressing visual concepts through code compared to previous generations like GPT-4.”