"Drawing" the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok (www.tryai.dev)

In an intriguing experiment, four AI models—GPT-5.6 Sol, Claude Fable 5, Grok 4.5, and Gemini 3.6 Flash—were tested for their drawing capabilities by recreating classic artworks like the Mona Lisa and Van Gogh's Starry Night. Each model was given a virtual drawing arena complete with a colored-pencil toolkit, allowing them to set colors, pressure, and stroke widths, and evaluate their work throughout the process. The goal was to assess model performance on both predetermined targets and open-ended prompts, resulting in a total of 28 drawings. Notably, GPT-5.6 Sol emerged as the most proficient, delivering high-quality outputs while being cost-effective, while Claude Fable 5, despite producing decent quality, incurred significant time and expense. This study sheds light on the capabilities and limitations of generative AI in creative tasks, highlighting how different models manage tool use, drawing techniques, and resource efficiency. It exposes the performance disparity between frontier and open-weight models, with Grok 4.5 struggling to produce usable artworks and highlighting the necessary evolution of AI models for better understanding of visual outputs. The findings not only contribute to ongoing discussions about AI creativity but also provide valuable insights for developers seeking to advance AI visual capabilities.