DefectBench (Building Facades) - Surface Stain Counting Error: leaderboard
Metric: Mean absolute error of the model's count of surface-stain instances per image, on DefectBench's 487 building-facade evaluation images (crack 117, material loss 102, surface stain 134, external fixings 122), zero-shot, JSON answers in one multi-turn dialogue per image; counts follow the model's own identified types; lower is better. Source: arxiv.org. Saturation forecast: Not forecast. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 32B Instruct | 0.94 | #443 |
| 2 | Qwen 3 VL 8B Instruct | 0.95 | #401 |
| 3 | GLM-4.5V | 0.98 | #339 |
| 4 | GLM-4.6V | 1.01 | #309 |
| 5 | Qwen 2.5 VL 72B Instruct | 1.03 | #364 |
| 6 | Gemini 3 Pro (Preview) | 1.06 | #64 |
| 7 | Qwen 3 VL 32B (Thinking) | 1.09 | #287 (Qwen 3 VL 32B) |
| 8 | Qwen 2.5 VL 7B Instruct | 1.1 | #643 |
| 9 | Qwen 3 VL 32B Instruct | 1.15 | #276 |
| 10 | GPT-5.2 Pro | 1.22 | #57 |
| 11 | Claude Opus 4.1 | 1.22 | #132 |
| 12 | Gemini 3 Flash (Preview) | 1.24 | #78 |
| 13 | GPT-4o | 1.34 | #333 |
| 14 | Seed 1.8 | 1.45 | #136 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=defectbench-building-facades-surface-stain-counting-error · How It Works · Data refreshed daily, snapshot 2026-10-11.