Pyyan / Compare / OWLv2 vs RF-DETR vs YOLO26 vs YOLOv12 vs RTMDet
OWLv2 vs RF-DETR vs YOLO26 vs YOLOv12 vs RTMDet
5 of 5Object Detection · verified 13 Aug 2026
| Specification | OWLv2 | RF-DETR | YOLO26 | YOLOv12 | RTMDet |
|---|---|---|---|---|---|
| Summary | Open-vocabulary detection trained with self-labelling. | First real-time detector past 60 mAP on COCO. | The latest in the line most people mean by object detection. | Attention-centric YOLO, still widely deployed. | Pure throughput, and an MIT licence. |
| COCO mAP | ~45 zero-shot | 60.5 | ~55 | ~55 | ~52 |
| Speed | Moderate | Real time | Very high | Very high | 300+ |
| Licence | Apache 2.0 | Apache 2.0 | AGPL-3.0 | AGPL-3.0 | MIT |
| Family | Open vocabulary | Transformer, set prediction | Single stage | Single stage | Single stage |
| Open vocabulary | Yes | No | No | No | No |
| Released | 2023 | 2025 | 2026 | 2025 | 2022 |
| Category | Object Detection | Object Detection | Object Detection | Object Detection | Object Detection |
| Official | Google ↗ | Roboflow ↗ | Ultralytics ↗ | Ultralytics ↗ | OpenMMLab ↗ |
Highlighted rows are where these differ.
OWLv2
- Accepts an image as the query, not just text
Best for querying by example image.
RF-DETR
- Leads the RF100-VL domain-transfer benchmark
- DINOv2 backbone, markedly better on occluded objects
- Apache 2.0, which YOLO is not
Best for most custom detection tasks.
YOLO26
- Strong on Jetson, Snapdragon and ARM CPUs
- AGPL-3.0: open-source your derivative work or buy a licence
Best for edge and mobile deployment.
YOLOv12
- Attention added to the classic single-stage design
- Same licensing consideration as YOLO26
Best for teams already on the YOLO toolchain.
RTMDet
- Wins on raw speed where accuracy is sufficient
- MIT, so no derivative-work obligation
Best for high frame-rate video.