Benchmark Results

Model Performance

Comprehensive evaluation of 10 embedding models across 9 diverse product datasets.

Models Evaluated

Datasets

57.6%

Best Avg P@1

Total Evaluations

Generic Models Leaderboard

General-purpose embedding models evaluated across all product datasets.

Rank	Model	Provider	Embed Dim	Input Res	Avg P@1	Avg P@5	Avg R@1	Avg R@5	Avg mAP@10
1	General V5.1	nyris	768	336	57.63%	43.19%	16.29%	41.33%	50.34%
2	PE-Core L/14	Meta	1024	336	42.87%	29.21%	12.42%	30.79%	34.04%
3	Vertex AI Multi-Modal	Google	1408	N/A	42.81%	29.68%	12.61%	31.50%	34.81%
4	SigLIP2 SO400M	Google	1152	384	41.16%	28.86%	10.09%	26.66%	31.22%
5	DINOv3 ViT-L/16	Meta	1024	224	40.23%	26.33%	11.44%	26.77%	29.80%
6	DINOv2 Large	Meta	1024	224	34.01%	21.53%	9.10%	20.89%	23.30%
7	Cohere Embed V4	Cohere	1024	N/A	33.87%	21.66%	12.73%	28.20%	28.55%
8	Jina Embeddings V4	Jina AI	768	Dynamic	26.27%	16.62%	7.11%	18.77%	18.08%
9	Nomic Embed MM 3B	Nomic AI	768	Dynamic	25.78%	15.90%	7.27%	17.88%	17.28%

Specialized models trained for specific product domains, evaluated only on their target datasets.

Model	Provider	Target Domain	Embed Dim	Input Res	P@1	P@5	R@1	R@5	mAP@10
Automotive V1	nyris	Specialized	768	336	32.49%	26.18%	11.13%	28.44%	33.96%

Loading retrieval results...

Query image

Correct match (same product)

Incorrect match (different product)