Which frontend skill setup actually makes better UI?
This benchmark gave the same app prompts to 15 skill profiles: every single skill, every pair, every triple, and the all-four combination. Each profile generated its own Next.js app. Then the outputs were built, screenshotted, and graded against the same rubric.
The winner was:
gpt-taste__frontend-skill__frontend-design
The interesting part is that the all-four setup did not win. More guidance helped up to a point, then it started to blur the design direction.
- Full results write-up
- Source notes for writing about the benchmark
- Screenshot gallery
- Contact-sheet comparisons
- Generated apps
- Rubric
- Prompt set
| Rank | Profile | Score | Evidence | Source | Read |
|---|---|---|---|---|---|
| 1 | gpt-taste__frontend-skill__frontend-design |
91 | screens | app | Best overall. Bold, but still controlled. |
| 2 | frontend-skill__design-taste-frontend__frontend-design |
90 | screens | app | Best no-GPT-Taste setup. Strong for product suites and operational UI. |
| 3 | frontend-design |
89 | screens | app | Best single skill. Focused direction beat several combinations. |
| 4 | frontend-skill__frontend-design |
89 | screens | app | Strong simple pair. Expressive without too many competing rules. |
| 5 | gpt-taste__frontend-skill__design-taste-frontend__frontend-design |
89 | screens | app | All four skills. Good, but not the winner. |
| 6 | gpt-taste__frontend-skill |
88 | screens | app | GPT-Taste works much better when paired with restraint. |
| 7 | frontend-skill__design-taste-frontend |
88 | screens | app | Best practical pair for dashboards and utility screens. |
| 8 | gpt-taste__frontend-skill__design-taste-frontend |
88 | screens | app | Solid product UI with a bolder first impression. |
| 9 | design-taste-frontend__frontend-design |
87 | screens | app | Structured and distinctive, but not as complete as the top group. |
| 10 | gpt-taste__design-taste-frontend |
87 | screens | app | A bolder operational hybrid. |
| 11 | gpt-taste__frontend-design |
86 | screens | app | Memorable, but less practical on dashboards. |
| 12 | design-taste-frontend |
86 | screens | app | Best single skill for dashboard-style work. |
| 13 | gpt-taste__design-taste-frontend__frontend-design |
86 | screens | app | Interesting, but uneven. |
| 14 | frontend-skill |
85 | screens | app | Clean and tasteful, but safer than the best combinations. |
| 15 | gpt-taste |
84 | screens | app | Strong hero instincts, weaker utility and mobile resilience. |
The best result came from combining a bold visual skill, a practical frontend skill, and a strong concept/design skill. The all-four setup was good, but slightly less decisive.
| Artifact | Count |
|---|---|
| Skill profiles | 15 |
| Generated apps | 15 |
| Passing builds | 15 / 15 |
| Individual screenshots | 210 |
| Contact sheets | 14 |

