◌
Same prompt
Every model receives the same frontend challenge brief, constraints, and output format.
AI frontend artifacts, curated side by side
Same challenge. Many AI works. No scores.
One frontend prompt is given to multiple AI models. VibeBench places their single-file HTML works side by side, so you can judge style, craft, and completeness yourself.
No scores. Just works. You judge.
18Challenges
64Model submissions
9AI models
100%Single-file HTML
Philosophy
Compare craft, not scores.
Frontend quality is contextual. VibeBench avoids compressing taste, structure, motion, and completion into a synthetic number.
▣
Single-file works
Each submission is a self-contained HTML file, including layout, styles, and interactions.
◐
Human judgment
VibeBench avoids synthetic scores. You inspect the works directly and decide what matters.
Challenges
Same brief, different instincts.
Select a frontend challenge to update the prompt, model runs, work previews, and comparison wall.
Models
Participants, presented evenly.
Model cards show objective run metadata and output tendencies. No recommendations, no rankings, no winners.
Works
Open the works. Feel the difference.
Each card represents a generated single-file HTML artifact with a curated preview and descriptive metadata.
Compare
Compare the works side by side.
We do not compress craft into a number. Look closely, switch context, and decide what matters to you.
Observation lenses
Ways to look, not scoring criteria.
These lenses are not scoring criteria. A designer, a developer, and a founder may value different things in the same work.
Recent traces
A living bench of artifacts.
Submission activity shown as a gentle path, not a leaderboard.
FAQ