VGBench: leaderboard
VGBench — generating and understanding vector graphics (SVG, TikZ, Graphviz); average score across formats.
Metric: Avg Score (SVG+TikZ+Graphviz). Source: vgbench.github.io. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 75 | #343 |
| 2 | GPT-4 | 73.5 | #438 |
| 3 | Qwen 2 72B | 70.7 | #521 |
| 4 | Gemini 1.5 Pro | 69.5 | #445 |
| 5 | Llama 3 70B | 64.7 | #631 |
| 6 | Qwen 2 7B | 59.8 | #1095 |
| 7 | GPT-3.5 Turbo | 58.7 | #865 |
| 8 | Llama 3 8B | 51.1 | #1242 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vgbench · How It Works · Data refreshed daily, snapshot 2026-10-20.