Roboflow Vision Evals (October 2026, four tasks without OCR)
Multimodal
Published incomplete aggregate for models without OCR results: unweighted mean of low-tier detection mAP@50 and judged counting, identification and reasoning. Kept separate from the full five-task aggregate.
Top models (higher is better)
| Model | Score |
|---|---|
| Muse Spark 1.3 | 74.7 |
| Mage-VL | 28.3 |