LetTokenBurnDesign Creativity Bench: Measuring creativity in LLM-Generated UI
Evaluates diversity and appropriateness of LLM-generated UI designs using 160 open-ended briefs, measuring originality, creative range, and adherence to acceptance criteria against human baselines.
$ npm i -g lettokenburn-cli$ ltb bench info design-creativity-bench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 65.5 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GPT-6.1 Sol | 63.7 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Grok-4.7 | 62.2 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Gemini 3.8 Flash | 60.6 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Qwen3.8 Max | 58.1 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GPT-5.6 Sol | 51.3 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Claude Opus 5.5 | 51.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Claude Fable 5.1 | 48.6 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Muse Spark 1.3 | 47.2 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| HY4 Preview | 46.2 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| DeepSeek V4.1 Flash | 43.6 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Grok-4.6 | 43.4 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Pareto 26.10 Preview | 42.5 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GLM-5.3 | 41.9 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
The benchmark uses 160 paired design briefs sharing goals but differing in domain/content to generate HTML UIs. It evaluates outputs via segment-level hybrid embeddings for similarity and checks against acceptance criteria.
Scoring combines originality (cross-model dissimilarity), creative range (intra-model distinctiveness), and appropriateness (% criteria met). LLMs show high appropriateness (>90%) but significantly lower originality and creative range than humans.