VectorGym - SVG Editing: leaderboard
Metric: SVG Editing task score (0-100): applying a human-written multi-step edit to an SVG, mean of the VLM-judge score, DINO similarity, 100 minus MSE and 100 minus LPIPS of the rendered output against the human-made target (all 0-100), on VectorGym's human-annotated test split (real-world SVGs from SVG-Stack); GPT-5.1 judges; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 88.71 | #77 |
| 2 | Claude Sonnet 4.5 | 88.07 | #138 |
| 3 | GPT-5.1 | 87.71 | #131 |
| 4 | GPT-4o | 82.35 | #333 |
| 5 | Gemini 2.5 Flash | 81.3 | #237 |
| 6 | Qwen 3 VL 235B A22B Instruct | 80.11 | #264 |
| 7 | Qwen 3 VL 8B Instruct | 77.89 | #401 |
| 8 | GLM-4.5V | 68.34 | #339 |
| 9 | Qwen 2.5 VL 32B Instruct | 59.61 | #443 |
| 10 | Qwen 2.5 VL 72B Instruct | 57.52 | #364 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=vectorgym-svg-editing · How It Works · Data refreshed daily, snapshot 2026-10-11.