VCG-Bench - Vision-to-XML - Style Consistency: leaderboard
Converting a raster diagram, with a Gemini-3-Pro structured caption, into editable draw.io mxGraph XML. A Gemini-3-Pro judge, itself one of the evaluated models, rates the rendered result against the original for visual style, layout and aesthetics, normalized to 0-1 with higher better; output that fails to render scores 0 (VCG-Bench, Su, Dong et al., HKUST (Guangzhou), 2026).
Metric: Style Consistency Score (0-1, Gemini-3-Pro judge). Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 0.78 | #135 |
| 2 | Gemini 3 Flash | 0.76 | #159 |
| 3 | Claude Opus 4.5 | 0.65 | #152 |
| 4 | Claude Sonnet 4.5 | 0.64 | #244 |
| 5 | GPT-5.2 | 0.56 | #172 |
| 6 | GLM-4.6V | 0.34 | #698 |
| 7 | Qwen 3 VL 235B A22B | 0.15 | #477 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vcg-bench-vision-to-xml-style-consistency · How It Works · Data refreshed daily, snapshot 2026-11-12.