The pelican benchmark is saturated so we made 9 models draw a MacBook Pro in SVG (playcode.io)

The pelican benchmark, once a rigorous test for AI drawing capabilities, has become ineffective as all contemporary models easily pass it due to its overexposure in training datasets and the lack of precise grading criteria. To address this, researchers introduced a new challenge: instructing AI models to draw a MacBook Pro 16 in SVG format. This task is significant because it presents a recognizable object with clear proportions and complex details, allowing for a more discerning evaluation of the models' artistic abilities. Nine leading models were tested at various effort levels, revealing that while the best result came from Claude Fable 5 at high effort ($0.73), Gemini 3 Flash delivered a simple yet coherent design for just $0.007. The benchmark highlighted critical insights about effort levels and their impact on output quality; while increasing effort improved results up to a point, it led to diminishing returns and failures at maximum settings. This test also exposed API errors not previously detected, emphasizing the importance of rigorous benchmarking in AI development. The introduction of a demanding and widely recognized object in this evaluation not only refines model assessment but also challenges developers to innovate further in drawing capabilities.