Verdict

RF-DETR is more accurate per parameter: at matched size on an RTX 5070 Ti it averages 3.9 mAP points higher than YOLOX and wins 3 of 4 pairs, and RF-DETR-L tops the range at 58.6 mAP. YOLOX is 32.2% faster on average in PyTorch and scales from 0.91M to 99.07M parameters. Pick by accuracy per parameter against throughput and range.

RF-DETR fields 4 measured variants and YOLOX fields 6, all on the same COCO val2017 protocol on an RTX 5070 Ti. RF-DETR's variants sit between 30.47M and 33.93M parameters, so all 4 pair against YOLOX-M (25.33M) by count.

ModelFamilyParams (M)mAP@50-95FPS
YOLOX-NanoYOLOX0.928.853.6
YOLOX-TinyYOLOX5.135.561.6
YOLOX-SYOLOX9.044.350.0
YOLOX-MYOLOX25.351.748.2
RF-DETR-NRF-DETR30.551.439.5
RF-DETR-SRF-DETR32.155.135.8
RF-DETR-MRF-DETR33.757.330.3
RF-DETR-LRF-DETR33.958.625.1
YOLOX-LYOLOX54.255.442.8
YOLOX-XYOLOX99.156.339.6
RF-DETR and YOLOX variant ladders interleaved by parameter count on NVIDIA RTX 5070 Ti, PyTorch FP32, batch 1. mAP in percent form. The family column shows where each frontier sits at every size.
Live chartverified data
Accuracy vs parameters on COCO val2017, RF-DETR and YOLOX variants highlighted against the full field.

Accuracy at matched compute

mAP is shown in percent. Each RF-DETR variant is paired with the nearest YOLOX variant by parameter count, which is YOLOX-M at 51.7 mAP for all 4. RF-DETR-S, RF-DETR-M and RF-DETR-L win by 3.4, 5.6 and 6.9 points; RF-DETR-N trails at 51.4. The mean edge is 3.9 points. At the top, RF-DETR-L reaches 58.6 mAP; YOLOX-X, far heavier at 99.07M, reaches 56.3. At the small end YOLOX goes down to YOLOX-Nano (0.91M, 28.8 mAP), a size RF-DETR does not field.

Speed

Averaged across the 4 pairs, YOLOX is 32.2% faster than RF-DETR in PyTorch. YOLOX-M runs 48.2 FPS; the RF-DETR variants it faces run between 39.5 and 25.1 FPS, slowing as accuracy rises.

Licensing
RF-DETR license
Apache-2.0
YOLOX license
Apache-2.0
Both families
permissive, cleared for commercial use

Which family to pick

Pick RF-DETR for accuracy per parameter: RF-DETR-S and up beat their YOLOX match by 3.4 points or more, and RF-DETR-L reaches 58.6 mAP. Pick YOLOX for throughput, 32.2% faster on average, and for small models down to 0.91M parameters. Both ship under Apache-2.0, so licensing does not force the call.

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.