Design Creativity Bench
Collection
Collection of Artifacts of the Design Creativity Bench by Kombai • 4 items • Updated • 12
Datasets:
model large_stringclasses 15
values | label large_stringclasses 15
values | lab large_stringclasses 10
values | effort large_stringclasses 3
values | prompt_id large_stringclasses 20
values | task large_stringclasses 10
values | domain large_stringclasses 2
values | archetype large_stringclasses 4
values | html large_stringlengths 11.6k 139k |
|---|---|---|---|---|---|---|---|---|
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task1_1 | 1 | 1 | form_flow | <!DOCTYPE html>
<html lang='en'>
<head>
<meta charset='UTF-8'>
<meta name='viewport' content='width=device-width, initial-scale=1.0'>
<meta name='theme-color' content='#f7f8f4'>
<title>Choose your plan · Grove</title>
<style>
:root{--page:#f7f8f4;--white:#fff;--ink:#20392e;--text:#37493f;--muted:#798279;--green:#285f45... |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task1_2 | 1 | 2 | form_flow | <!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta name="theme-color" content="#244b38">
<meta name="description" content="Choose a cloud monitoring or automated deployment plan for your Stratum workspace.">
<title>Choose... |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task2_1 | 2 | 1 | detail_view | "<!DOCTYPE html>\n<html lang='en'>\n<head>\n<meta charset='UTF-8'>\n<meta name='viewport' content='w(...TRUNCATED) |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task2_2 | 2 | 2 | detail_view | "<!DOCTYPE html>\n<html lang='en'>\n<head>\n<meta charset='UTF-8'>\n<meta name='viewport' content='w(...TRUNCATED) |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task3_1 | 3 | 1 | form_flow | "<!DOCTYPE html>\n<html lang='en'>\n<head>\n<meta charset='UTF-8'>\n<meta name='viewport' content='w(...TRUNCATED) |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task3_2 | 3 | 2 | form_flow | "<!DOCTYPE html>\n<html lang='en'>\n<head>\n<meta charset='UTF-8'>\n<meta name='viewport' content='w(...TRUNCATED) |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task4_1 | 4 | 1 | form_flow | "<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" cont(...TRUNCATED) |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task4_2 | 4 | 2 | form_flow | "<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" cont(...TRUNCATED) |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task5_1 | 5 | 1 | landing_page | "<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" cont(...TRUNCATED) |
gpt_6_astra_max | GPT-6 Astra | OpenAI | max | task5_2 | 5 | 2 | landing_page | "<!DOCTYPE html>\n<html lang='en'>\n<head>\n<meta charset='UTF-8'>\n<meta name='viewport' content='w(...TRUNCATED) |
End of preview. Expand in Data Studio
Design Creativity Bench
Public subset of the data for Design Creativity Bench: Measuring creativity in LLM-Generated UI. Evaluation code: the design-creativity-bench repository accompanying the paper.
The benchmark measures originality (distinctiveness among models on the same prompt), creative range (how much a model's designs change between two prompts for the same UI goal in different product domains) and appropriateness (share of acceptance criteria met).
Configs
| Config | Rows | Contents |
|---|---|---|
designs |
300 | One self-contained HTML page per (model, prompt): 15 models x 20 public prompts. Columns: model, label, lab, effort, prompt_id, task, domain, archetype, html |
prompts |
20 | The public prompts (10 tasks x 2 product domains) with vertical, the prompt text, and the acceptance criteria as a JSON string |
models |
15 | Panel models, their lab and reasoning effort |
paper_scores |
16 | The paper's published results, panel and human reference |
subset_scores |
15 | Originality, creative range and diversity of the panel on the 20 public prompts |
ladders |
63 | Controlled-ladder studies: anchor/seed pages, nine variants per study, and each variant's edit specification |
paper_ladders |
36 | The paper's ladder results |
generation/ holds what the models were sent: system prompt, user-prompt wrapper, tool schemas, follow-up messages and per-model settings.
Subset scores are measured on the 20 public prompts and are not comparable with the paper's results, which use the benchmark's own prompts.
Related datasets
- Precomputed features and embeddings for
designs:Kombai/design-creativity-bench-features - Projection training data:
Kombai/design-creativity-bench-projection
License
MIT. Designs are outputs of commercially licensed model APIs.
Citation
@misc{rusia2026design,
author = {Rusia, Aman and Bhole, Abhijit and Gupta, Prashank and Dey, Dipanjan},
title = {Design Creativity Bench: Measuring creativity in LLM-Generated UI},
year = {2026},
publisher = {arXiv},
doi = {10.48550/ARXIV.2610.11539},
url = {https://arxiv.org/abs/2610.11539}
}
- Downloads last month
- 63