Model Accuracy#

The following two tables present the absolute accuracy drop calculated as the accuracy difference between OV-accuracy and the original frame work accuracy for FP32, and the same for INT8, BF16 and FP16 representations of a model on three platform architectures. The third table presents the GenAI model accuracies as absolute accuracy values. Please also refer to notes below the table for more information.

A - Intel® Core™ Ultra 9-185H (AVX2), INT8 and FP32
B - Intel® Xeon® 6338, (VNNI), INT8 and FP32
C - Intel® Xeon 6972P (VNNI, AMX), INT8, BF16, FP32
D - Intel® Arc-B580, INT8 and FP16

Model Accuracy for INT8#
OpenVINO™ Model name	dataset	Metric Name	A, INT8	B, INT8	C, INT8	D, INT8
bert-base-cased	SST-2_bert_cased_padded	spearman@cosine	2.57%	2.65%	2.95%	2.63%
Detectron-V2	COCO2017_detection_91cl_bkgr	coco_orig_precision
mobilenet-v2	ImageNet2012	accuracy @ top1	-0.91%	-0.93%	-0.91%	-1.03%
resnet-50	ImageNet2012	accuracy @ top1	-0.12%	-0.12%	-0.15%	-0.15%
ssd-resnet34-1200	COCO2017_detection_80cl_bkgr	map	0.00%	0.00%	0.03%	0.07%
yolo_v11	COCO2017_detection_80cl	AP@0.5:0.05:0.95

Model Accuracy for BF16, FP32 and FP16 (FP16: Arc only. BF16: Xeon® 6972P only)#
OpenVINO™ Model name	dataset	Metric Name	A, FP32	B, FP32	C, FP32	D, FP16
bert-base-cased	SST-2_bert_cased_padded	spearman@cosine	0.00%	0.00%	0.00%	0.00%
Detectron-V2	COCO2017_detection_91cl_bkgr	coco_orig_precision
mobilenet-v2	ImageNet2012	accuracy @ top1	0.00%	0.00%	0.02%	0.02%
resnet-50	ImageNet2012	accuracy @ top1	0.00%	0.00%	0.00%	-0.01%
ssd-resnet34-1200	COCO2017_detection_80cl_bkgr	map	0.02%	0.02%	0.02%	-0.23%
yolo_v11	COCO2017_detection_80cl	AP@0.5:0.05:0.95	0.03%	-2.21%	-2.21%	-2.21%

Model Accuracy for AMX-FP16, AMX-INT4, Arc-FP16 and Arc-INT4 (Arc™ B-series)#
OpenVINO™ Model name	dataset	Metric Name	A, AMX-FP16	B, AMX-INT4	C, Arc-FP16	D, Arc-INT4
DeepSeek-R1-Distill-Llama-8B	Data Default WWB	Similarity	98.1%	94.4%	99.5%	94.0%
DeepSeek-R1-Distill-Qwen-1.5B	Data Default WWB	Similarity	96.5%	92.4%	99.7%	92.3%
Gemma-3-4B-it	Data Default WWB	Similarity	92.0%	83.9%		84.9%
Llama-2-7B-chat	Data Default WWB	Similarity	99.3%	93.4%	99.8%	93.4%
Llama-3-8B	Data Default WWB	Similarity	98.8%	94.3%	99.7%	94.5%
Llama-3.2-3b-instruct	Data Default WWB	Similarity	97.9%	94.2%	99.7%	94.1%
Phi4-mini-instruct	Data Default WWB	Similarity	89.1%	92.1%	99.5%	92.4%
Qwen2-VL-7B	Data Default WWB	Similarity	97.5%	88.1%	99.8%	91.4%
Qwen3-8B	Data Default WWB	Similarity	97.8%	92.3%		93.0%
Flux.1-schnell	Data Default WWB	Similarity	95.4%	96.1%		92.1%
Stable-Diffusion-V1-5	Data Default WWB	Similarity	96.3%	93.3%	99.5%	93.7%

Notes: For all accuracy metrics a “-”, (minus sign), indicates an accuracy drop. The Similarity metric is the distance from “perfect” and as such always positive. Similarity is cosine similarity - the dot product of two vectors divided by the product of their lengths.