At matched compute on an RTX 5070 Ti, neither family wins outright. YOLOv9 is more accurate per parameter at the small end and 22.8% faster on average in PyTorch; RF-DETR pulls ahead on accuracy at the large end, topping out at 58.6 mAP against YOLOv9's 57.1. Across 4 nearest-parameter pairs the accuracy split is even, 2 to 2, with YOLOv9 averaging 1.5 mAP points higher overall.
RF-DETR fields 4 measured variants; YOLOv9 fields 4. Both run the same COCO val2017 protocol at 640 px on an RTX 5070 Ti, so the two ladders align by parameter count.
| Model | Family | Params (M) | mAP@50-95 | FPS |
|---|---|---|---|---|
| YOLOv9-T | YOLOv9 | 2.0 | 4178.0 | 31.9 |
| YOLOv9-S | YOLOv9 | 7.2 | 5045.0 | 30.8 |
| YOLOv9-M | YOLOv9 | 20.1 | 5610.0 | 37.1 |
| YOLOv9-C | YOLOv9 | 25.5 | 5709.0 | 42.3 |
| RF-DETR-N | RF-DETR | 30.5 | 5135.0 | 39.5 |
| RF-DETR-S | RF-DETR | 32.1 | 5512.0 | 35.8 |
| RF-DETR-M | RF-DETR | 33.7 | 5732.0 | 30.3 |
| RF-DETR-L | RF-DETR | 33.9 | 5855.0 | 25.1 |
Accuracy at matched compute
mAP is shown in percent. Pairing by nearest parameter count, the two families split the 4 pairs 2 and 2, and YOLOv9 averages 1.5 mAP points higher overall. The result depends on size class: YOLOv9-C leads the lower pairs, while RF-DETR-L reaches 58.6 mAP at the top, above YOLOv9-C's 57.1.
Speed
YOLOv9 is faster at matched compute: 22.8% faster on average in PyTorch FP32. RF-DETR's most accurate variant, RF-DETR-L, runs 25.1 FPS against YOLOv9-C at 42.3.
Where the frontier crosses
The accuracy lead changes hands with model size. At the small end YOLOv9 leads: its parameter-efficient variants sit above RF-DETR's smallest, RF-DETR-N, which needs 30.47M parameters to reach 51.4 mAP. At the large end RF-DETR leads, with RF-DETR-L at 58.6 mAP. There is no single winner; the choice follows your size class.
- RF-DETR
- Apache-2.0 (permissive)
- YOLOv9
- MIT (permissive)
- Commercial use
- Both families ship under permissive licenses
Which family to pick
Pick YOLOv9 for small and mid-size deployments and for raw throughput: it is faster on average and more parameter-efficient at the small end. Pick RF-DETR when peak accuracy matters and you can run a larger model, since it reaches 58.6 mAP at the top of the measured range. Both ship under permissive licenses. See the per-variant specs on the RF-DETR and YOLOv9 model pages.
Every number on this page comes from the verified dataset: same 500-image COCO val2017 slice, conf 0.001, IoU 0.6, max 300 detections, pycocotools mAP, identical protocol across all hardware and runtimes. The full protocol is on the methodology page. To rerun this comparison with your own filters, open compare. Accuracy is measured on LibreYOLO retrained checkpoints; other weight sources can yield different values.