hailo

Hailo-8

Workhorse 26-TOPS NPU on M.2. The default low-power vision accelerator for 1080p multi-stream perception.

top models on this chip

Object detection

Sorted by latency p50 · matched-pair runs on Hailo-8

Object detection

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	YOLO26-Nano	INT8	8.28 ms	121 FPS	97.5%	2.3 W
2	YOLO27-Edge	INT8	21.2 ms	47 FPS	97.5%	2.3 W
3	Hailo-YOLO27-INT8	INT8	22.7 ms	44 FPS	98.7%	2.2 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Hailo-8

Depth

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	Depth-Anything-Edge	INT8	40.9 ms	24 FPS	98.6%	2.3 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Hailo-8

Embedded ASR

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	Moonshine-Embedded	INT8	342.7 ms	19 tok/s	98.2%	2.1 W
2	Whisper-Edge-Tiny	INT8	355.9 ms	18 tok/s	98.1%	2.1 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Hailo-8

Re-ID

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	OSNet-AIN	INT8	8.88 ms	113 FPS	98.5%	2.3 W
2	ArcFace-Edge	INT8	25.5 ms	39 FPS	98.3%	2.3 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Hailo-8

Tracking

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	StrongSORT-Mobile	INT8	8.25 ms	121 FPS	98.6%	2.1 W
2	ByteTrack-Edge	INT8	8.41 ms	119 FPS	97.3%	2.1 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Hailo-8

Pose

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	RTMPose-Edge	INT8	15.7 ms	64 FPS	97.6%	2.1 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Hailo-8

OCR

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	PaddleOCR-Edge-v5	INT8	12.0 ms	83 FPS	97.3%	2.4 W