OpenUI Generative UI Benchmark: leaderboard
OpenUI gives a model a brief listing the requirements for one screen and has it emit a component graph, four generations per brief across 46 briefs in five complexity bands. Each result is checked by the SDK's own parser rather than a judge: it must parse, have a root, resolve every reference, invent no components, keep props valid and not be truncated. The number is the share that passes, higher is better.
Metric: Structural Validity (%). Source: www.openui.com. Saturation forecast: Around January 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 99.5 | #19 |
| 2 | Grok 4.6 | 99.5 | #26 |
| 3 | Claude Opus 4.8 | 98.9 | #37 |
| 4 | Gemini 3.7 Flash | 98.9 | #23 |
| 5 | GPT-5.6 Terra | 98.4 | #41 |
| 6 | Claude Sonnet 5 | 98.4 | #57 |
| 7 | Kimi K3 | 96.2 | #30 |
| 8 | Claude Opus 5 | 96.2 | #12 |
| 9 | Muse Spark 1.2 | 96.2 | #62 |
| 10 | Claude Sonnet 4.6 | 92.9 | #90 |
| 11 | GLM-5.3 | 90.8 | #39 |
| 12 | Inkling Small | 88.6 | #138 |
| 13 | DeepSeek V4 Flash | 85.8 | #137 |
| 14 | DeepSeek V4 Pro | 84.2 | #102 |
| 15 | GPT-5.6 Luna | 83.7 | #68 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openui-generative-ui-benchmark · How It Works · Data refreshed daily, snapshot 2026-10-20.