Vision trends · updated Oct 2, 2026

Which open-weight vision models get downloaded most

A daily read of Hugging Face's top 5,000 models, focused on vision models such as image classification, object detection and segmentation: the most downloaded, new uploads in the top 5,000, what they are used for, and who makes them.

Most downloaded

The most-downloaded open-weight vision models

Ranked by downloads over the last 30 days. Fine-tuned versions count; most format conversions and re-uploads are left out. Counts include automated fetches, which lift models that tools and test suites pull by default. How models are placed is explained under About this data.

#	Model	Maker	Task	Licence	Downloads	Valkyrie
1	clip-vit-base-patch32	openai	Zero-shot image classification	-	21.4M
2	mobilenetv3_small_100.lamb_in1k	timm	Image classification	Apache-2.0	21.3M	Fine-tune one like it →Fine-tune one →
3	vit-base-patch16-224	google	Image classification	Apache-2.0	10.0M	Fine-tune one like it →Fine-tune one →
4	clip-vit-large-patch14	openai	Zero-shot image classification	-	8.9M
5	siglip2-base-patch16-256	google	Zero-shot image classification	Apache-2.0	3.9M	Fine-tune one like it →Fine-tune one →
6	CLIP-ViT-B-32-laion2B-s34B-b79K	laion	Zero-shot image classification	MIT	3.8M	Fine-tune one like it →Fine-tune one →
7	CLIP-ViT-L-14-laion2B-s32B-b82K	laion	Zero-shot image classification	MIT	3.1M	Fine-tune one like it →Fine-tune one →
8	dinov2-small	facebook	Image features	Apache-2.0	2.8M	Fine-tune one like it →Fine-tune one →
9	dinov2-base	facebook	Image features	Apache-2.0	2.8M	Fine-tune one like it →Fine-tune one →
10	Depth-Anything-V2-Small-hf	depth-anything	Depth estimation	Apache-2.0	2.7M	Fine-tune one like it →Fine-tune one →
11	fashion-clip	patrickjohncyh	Zero-shot image classification	MIT	2.2M	Fine-tune one like it →Fine-tune one →
12	clip-vit-large-patch14-336	openai	Zero-shot image classification	-	2.2M
13	nsfw_image_detection	Falconsai	Image classification	Apache-2.0	2.2M
14	sam3	facebook	Segmentation	Custom licence	2.2M
15	siglip2-giant-opt-patch16-384	google	Zero-shot image classification	Apache-2.0	2.1M	Fine-tune one like it →Fine-tune one →
16	resnet18.a1_in1k	timm	Image classification	Apache-2.0	2.0M	Fine-tune one like it →Fine-tune one →
17	owlv2-base-patch16-ensemble	google	Zero-shot object detection	Apache-2.0	2.0M	Fine-tune one like it →Fine-tune one →
18	siglip-base-patch16-224	google	Zero-shot image classification	Apache-2.0	2.0M	Fine-tune one like it →Fine-tune one →
19	GLM-OCR	zai-org	Image to text	MIT	1.8M	Fine-tune one like it →Fine-tune one →
20	tf_efficientnetv2_s.in21k_ft_in1k	timm	Image classification	Apache-2.0	1.8M	Fine-tune one like it →Fine-tune one →
21	vit-base-patch16-224-in21k	google	Image features	Apache-2.0	1.8M	Fine-tune one like it →Fine-tune one →
22	blip-image-captioning-base	Salesforce	Image to text	BSD-3-Clause	1.6M	Fine-tune one like it →Fine-tune one →
23	siglip2-base-patch16-224	google	Zero-shot image classification	Apache-2.0	1.6M	Fine-tune one like it →Fine-tune one →
24	resnet50.a1_in1k	timm	Image classification	Apache-2.0	1.6M	Fine-tune one like it →Fine-tune one →
25	grounding-dino-base	IDEA-Research	Zero-shot object detection	Apache-2.0	1.5M	Fine-tune one like it →Fine-tune one →

New releases

New vision models in the top 5,000

Uploaded to Hugging Face in the last six months with at least 25 likes or a trending score of 10, ranked by downloads over the last 30 days. Most format conversions and re-uploads are left out.

tipsv2-b14googleZero-shot image classification · uploaded Apr 9, 2026172K downloads in 30 daysFine-tune one like it →Fine-tune one →

PP-OCRv6_medium_detPaddlePaddleImage to text · uploaded Jun 10, 202678K downloads in 30 daysFine-tune one like it →Fine-tune one →

PP-OCRv6_medium_recPaddlePaddleImage to text · uploaded Jun 10, 202666K downloads in 30 daysFine-tune one like it →Fine-tune one →

NuExtract3numindImage to text · uploaded Apr 29, 202634K downloads in 30 daysFine-tune one like it →Fine-tune one →

LeVJEPA-VideoMix-Largegalilai-groupImage features · uploaded Aug 27, 202629K downloads in 30 days

What they're used for

Downloads by task

Share of downloads of vision models in Hugging Face's top 5,000 on Oct 2, 2026, leaving out most format conversions and re-uploads.

Zero-shot image classification36.1%

Image classification32.2%

Image features10.0%

Image to text5.4%

Segmentation5.3%

Object detection3.8%

Depth estimation3.0%

Zero-shot object detection2.8%

Video classification0.7%

Keypoint detection0.6%

Who's winning

Share of downloads by maker

Each repo counts under the account that uploaded it, leaving out most format conversions and re-uploads. Change is in percentage points since Jul 4, 2026.

timm25.1% (−0.4 pts)

openai18.7% (−1.3 pts)

google16.5% (+7.3 pts)

facebook8.7% (+0.2 pts)

laion5.5% (+2.3 pts)

microsoft2.6% (−0.8 pts)

depth-anything2.4% (+0.6 pts)

PaddlePaddle1.9% (+0.3 pts)

Other18.6% (−8.2 pts)

Licences of the most-downloaded vision models

Apache-2.0 57.9%MIT 12.5%Other or none 29.6%

As declared on each repo, leaving out most format conversions and re-uploads.