apple

Apple Neural Engine (M4)

Reference target for on-device foundation-model workloads; benchmarks via Core ML.

top models on this chip

Object detection

Sorted by latency p50 · matched-pair runs on Apple Neural Engine (M4)

Object detection

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	YOLO27-Edge	FP16	19.1 ms	52 FPS	99.9%	7.1 W
2	RT-DETR-Edge	FP16	19.6 ms	51 FPS	99.8%	6.8 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Apple Neural Engine (M4)

Depth

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	MiDaS-Distilled-S	FP16	15.9 ms	63 FPS	99.4%	7.4 W
2	ZoeDepth-Mobile	FP16	22.7 ms	44 FPS	99.5%	6.4 W
3	Depth-Anything-Edge	MIXED	29.2 ms	34 FPS	98.7%	6.9 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Apple Neural Engine (M4)

Segmentation

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	Mask2Former-Mobile	FP16	36.3 ms	28 FPS	99.6%	7.3 W
2	SAM-3-Distilled	FP16	55.9 ms	18 FPS	99.5%	7.7 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Apple Neural Engine (M4)

Embedded ASR

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	Whisper-Edge-Tiny	INT8	203.5 ms	31 tok/s	97.7%	6.8 W
2	Distil-Whisper-Tiny	FP16	333.5 ms	19 tok/s	99.8%	7.3 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Apple Neural Engine (M4)

Multimodal

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	MobileCLIP-S2	FP16	34.9 ms	29 FPS	99.6%	6.5 W
2	MobileCLIP-B	FP16	66.9 ms	15 FPS	99.9%	6.6 W
3	Florence-2-Edge	FP16	70.2 ms	14 FPS	99.8%	7.2 W
4	DINO-v3-Distilled	FP16	78.1 ms	13 FPS	99.8%	7.1 W
5	SmolVLM-3B	MIXED	342.2 ms	2.9 FPS	99.4%	7.3 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Apple Neural Engine (M4)

Pose

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	MoveNet-Lightning	INT8	4.65 ms	215 FPS	98.2%	7.1 W
2	MediaPipe-Hands-Edge	INT8	5.29 ms	189 FPS	97.6%	6.2 W

top models on this chip

Sorted by latency p50 · matched-pair runs on Apple Neural Engine (M4)

OCR

#	Model	Quant	Latency p50	Throughput	Acc.	Power
1	PaddleOCR-Edge-v5	FP16	10.2 ms	98 FPS	99.4%	7.2 W
2	DocLayout-Edge	FP16	16.6 ms	60 FPS	99.7%	6.6 W