Are AI labs pelicanmaxxing?

This article examines a deep-dive investigation by Dylan Castillo into whether AI labs deliberately trained models to draw pelicans riding bicycles. The study used 8 animals × 6 vehicles = 48 prompts, run three times each through 7 models including GPT-5.6 Terra, Claude Sonnet 5, Gemini 3.5 Flash, and others. Results were evaluated with additional AI tools. The conclusion found no evidence of 'pelimaxxing'—pelicans on bicycles were not drawn better than other animals or vehicles, and no lab showed significant bias toward this specific combination. The article is a technical analysis of AI model behavior and benchmarking methodology.