"Drawing" the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok

# Summary Researchers created a "drawing arena" where four advanced AI models (GPT-5.6 Sol, Claude Fable 5, Grok 4.5, and Gemini 3.6 Flash) were given digital colored-pencil tools to reproduce the Mona Lisa and Starry Night or draw from text prompts, resulting in 28 total drawings. The experiment revealed significant capability gaps between frontier models, with Claude Fable 5 producing inferior output despite higher costs and longer processing times, while Grok 4.5 struggled with the basic task and open-weight models largely failed. These open-ended drawing tasks serve as a more revealing test of real-world model capability than traditional benchmarks, exposing trade-offs in cost, speed, and quality across different models.