skill significantly outperform previous iterations in visual frontend design. This isn't just about cleaner code; it's about an intentional shift toward aesthetic defaults. Testing these claims across a
reveals a model that prioritizes a professional, dark-themed aesthetic. However, a pattern emerges quickly: GPT-5.4 is remarkably consistent, perhaps to a fault. Whether prompted for a
frontend-design skill, takes a radically different approach. Its system prompts explicitly forbid "generic overused font families" and "clichéd color schemes." In practice, this results in designs that feel more human and less like a template. While it sometimes chooses eccentric fonts that lack proper padding, its adherence to the original spirit of a prompt—like capturing the "blocky" feel of
remains the dark horse in the design race. For months, it has maintained a reputation for solid visual outputs, and these tests confirm its status. In a redesign challenge for