11 Models

Embedding Models

State-of-the-art visual embedding models from leading providers, evaluated for product search performance.

Total Models

Open Source

Commercial

In-House

Generic Models

General-purpose embedding models evaluated across all benchmark datasets.

Rank	Model	Accessibility	Modality	Embedding Size	Input Size	Avg. R@1	Avg. R@5	Avg. mAP@20
1	GEM v5.1 (ours)	In-House	Vision	768	336	56.84%	72.35%	48.88%
2	DINOv3 ViT-L/16	Open Source	Vision	1024	224	40.58%	58.35%	29.73%
3	SigLIP2 SO400M	Open Source	Vision	1152	384	40.54%	56.81%	32.98%
4	PE-Core L/14	Open Source	Vision	1024	336	40.39%	57.42%	32.40%
5	Gemini Embedding 2	Commercial	Multi-Modal	3072	N/A	38.87%	56.36%	31.92%
6	Vertex AI Multi-Modal	Commercial	Multi-Modal	1408	N/A	38.67%	56.51%	32.14%
7	Cohere Embed v4	Commercial	Multi-Modal	1536	N/A	33.67%	46.76%	27.07%
8	DINOv2 Large	Open Source	Vision	1024	224	31.53%	45.50%	21.47%
9	Jina Embeddings v4	Open Source	Multi-Modal	2048	Dynamic	27.60%	38.70%	19.86%
10	Nomic Embed MM 3B	Open Source	Multi-Modal	2048	Dynamic	27.17%	38.48%	18.73%

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

Model	Accessibility	Target Domain	Embedding Size	Input Size	R@1	R@5	mAP@20
AEM v1 (ours)	In-House	Automotive	768	336	32.49%	51.60%	35.68%

Open Source

Freely available model weights that can be deployed on your own infrastructure. Offers flexibility and control over the inference pipeline.

Commercial

Third-party models accessed via API or commercial license (e.g. Vertex AI, Cohere). Typically hosted and maintained by the provider.

In-House

Models developed and operated in-house (e.g. GEM, AEM). Optimized for specific product search and domain workflows.

N/A

Input resolution is not available from the model's official documentation or specifications.

Dynamic

The input size is dynamically adjusted based on the model's resizing scheme and image resolution.