1D-Bench (Single-Round): leaderboard
Metric: Final score (0-100) for single-pass generation of an executable React project from a reference UI rendering and an exported design representation that may contain extraction defects, using a fixed React scaffold and a file-writing tool; an execution harness builds and renders the project, and a composite LPIPS-based visual similarity (with SSIM, text and layout cues, 0-100) is computed against the reference rendering; the final score is mean similarity over rendered instances times the rendering success rate; the success rates fit whole counts of 44 instances or a multiple of 44 (38 or a multiple for Qwen3-VL-235B-A22B-Instruct) and none fits the 204-instance evaluation set the paper describes; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 79.6 | #77 |
| 2 | Claude Sonnet 4.5 | 74 | #138 |
| 3 | Qwen 3 VL 235B A22B Instruct | 59 | #264 |
| 4 | GPT-5.2 | 49.8 | #105 |
| 5 | GLM-4.6V | 3.7 | #309 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=1d-bench-single-round · How It Works · Data refreshed daily, snapshot 2026-10-11.