Roboflow Vision Evals - Visual Understanding: leaderboard
Roboflow Vision Evals benchmark for visual QA tasks such as reading text from photos, counting objects, spotting defects, and understanding documents.
Metric: Pass Rate (%). Source: playground.roboflow.com. Status: years away from saturation. 77 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 35B A3B | 79.1 |
| 2 | Claude Fable 5 | 79.1 |
| 3 | Gemini 3.5 Flash | 79.1 |
| 4 | GPT-5.5 | 77.61 |
| 5 | GPT-5.4 | 77.61 |
| 6 | GPT-5.4 Mini | 77.61 |
| 7 | Qwen 3.5 122B A10B | 76.12 |
| 8 | GPT-5.6 Sol | 76.12 |
| 9 | GPT-5.6 Terra | 76.12 |
| 10 | Gemini 3.1 Pro (Preview) | 75.76 |
| 11 | Gemini 3 Flash | 74.63 |
| 12 | Qwen 3.5 Plus | 74.63 |
| 13 | GPT-5 Mini | 73.13 |
| 14 | Qwen 3.5 9B | 71.64 |
| 15 | Qwen 3.5 27B | 71.64 |
Interactive version: theaggregate.ai/benchmark?slug=roboflow-vision-evals-visual-understanding · How It Works · Data refreshed daily, snapshot 2026-09-05.