Are AI labs pelicanmaxxing?

This article examines a deep-dive analysis by Dylan Castillo investigating whether AI labs deliberately train models to generate images of pelicans riding bicycles. The study tests 7 models (GPT-5.6 Terra, Claude Sonnet 5, Gemini 3.5 Flash, Grok 4.5, Qwen3.7-Max, GLM-5.2, DeepSeek V4 Pro) across 48 prompts (8 animals × 6 vehicles) run three times each. Results show no evidence of 'pelicanmaxxing': pelicans on bicycles are not drawn better than other animals or vehicles, and no lab shows significant improvement for the specific combination. The conclusion is that the phenomenon is not supported by the data.