"Drawing" the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok
Four frontier models, a blank canvas, and colored pencils. We tracked every stroke, dollar, and output as they tried to draw the Mona Lisa.
AI Summary
Four frontier AI models—GPT-5.6, Claude, Gemini, and Grok—were given a blank canvas and colored-pencil tools to draw the Mona Lisa and other targets, with their strokes, costs, and outputs tracked. The models could set color, tip width, pressure, lay down strokes, smudge, erase, and view their own canvas to decide what to fix. The experiment ran 28 drawings across two objective targets (Mona Lisa and Starry Night) and five open-ended prompts.