Frontier AI Art Appraisal Test Reveals Gap
Experiment tested 4 frontier multimodal models on valuing 15 paintings worth $1.46B using image-only or image + metadata. Found 'recognition vs commitment gap': models recognize art visually but hesitate on valuation without text. Gemini 3.1 Pro led both settings; GPT-5.4 surged with metadata.




