DrawAI-Bench - Fidelity - Table: leaderboard
DrawAI-Bench (arXiv 2608.00548v1, Table 2): coding agents convert 80 raster diagrams, posters, slides and scientific figures into editable SVG with the DrawAI-Flow workflow, scored by deterministic rules and VLM rubric questions; each model appears under the strongest harness observed for it. Asset-level fidelity: judged grid structure, cell content ownership and readability of tables; 0-100, higher is better.
Metric: Fidelity score (0-100; rules and VLM rubric, strongest observed harness). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 98.2 | #22 |
| 2 | Claude Opus 4.8 | 96.3 | #39 |
| 3 | GPT-5.5 | 94.4 | #33 |
| 4 | Kimi K3 | 94.4 | #35 |
| 5 | Claude Fable 5 | 94.4 | #20 |
| 6 | GPT-5.4 | 85.2 | #83 |
| 7 | MiniMax-M3 | 79.6 | #151 |
| 8 | Gemini 3.5 Flash | 79.6 | #66 |
| 9 | Kimi K2.7 Code | 68.5 | #95 |
| 10 | MiMo-V2.5 | 61.1 | #196 |
| 11 | Gemini 3.1 Pro (Preview) | 57.4 | #64 |
| 12 | Qwen 3.7 Plus | 57.4 | #98 |
| 13 | MiMo-V2.5-Pro | 51.9 | #167 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=drawai-bench-fidelity-table · How It Works · Data refreshed daily, snapshot 2026-11-05.