Deconstructing Real World Creativity.
Exploring the frontier of AI design capabilities through real-world creative tasks, human references, and model comparisons.
How the models rank
How frontier models rank against professional human references on real design tasks.
Behind every finished work is a series of creative decisions
Frontier Design Benchmark compares professional human references with AI-generated outputs across real-world design tasks, helping define what models understand, what they miss, and how their creative capabilities stand today.
- 28
- human references
- 7
- frontier models
- 8
- design domains
- 196
- ranked outputs
Golden Gallery
Real design tasks. Human references. Model outputs.
Read the research
Paper
Read paper
Methodology
Methodology
Dataset
View sample
Frontier Design Benchmark: measuring AI on real-world design tasks
Contra Labs · October 2026 · 24 pages
How tasks are set, judged and ranked
How each brief is written, how designers compare outputs blind, and how pairwise votes become a ranking.
Data sample
A sample of tasks, references and model outputs for labs and data teams.